@fugood/buttress-server 2.24.2 → 2.25.0-beta.6

package/README.md

@@ -157,6 +157,25 @@ Examples:
   bricks-buttress --test-caps ggml-stt --test-caps-model-id BricksDisplay/whisper-ggml:ggml-small.bin
 ```
 
+## Compatibility Endpoints (Experimental)
+
+The server can expose OpenAI- and Anthropic-compatible HTTP endpoints in addition to the native RPC. Each endpoint is opt-in via the TOML config:
+
+```toml
+[openai_compat]
+enabled = true
+# cors_allowed_origins = "*"          # Or a list of origins; defaults to disabled
+
+[anthropic_messages]
+enabled = true
+# cors_allowed_origins = ["http://localhost:3000"]
+```
+
+| Endpoint              | Config flag                          |
+| --------------------- | ------------------------------------ |
+| `/oai-compat/v1/*`    | `[openai_compat] enabled = true`     |
+| `/anthropic-messages` | `[anthropic_messages] enabled = true` |
+
 ## Session State Cache
 
 The server supports session state caching for ggml-llm generators, which saves KV cache state to disk after completions. This enables:

package/config/sample.toml

@@ -17,9 +17,15 @@ log_level = "info"
 # max_body_size = "100MB"  # Supports string (e.g., "100MB", "1GB") or number in bytes
 
 [openai_compat]
+# enabled = true
 # cors_allowed_origins = ["http://localhost:3000", "https://example.com"]  # Restrict to specific origins
 # cors_allowed_origins = "*"  # Allow all origins (default)
 
+[anthropic_messages]
+# enabled = true
+# cors_allowed_origins = ["http://localhost:3000", "https://example.com"]
+# cors_allowed_origins = "*"
+
 [runtime]
 cache_dir = "./.buttress-cache"
 # huggingface_token = "hf_xx"

package/lib/index.d.mts

@@ -229,9 +229,11 @@ type GeneratorConfig = {
 type GlobalConfig = {
   runtime?: RuntimeConfig;
   openai_compat?: {
+    enabled?: boolean;
     cors_allowed_origins?: string | string[];
   };
   anthropic_messages?: {
+    enabled?: boolean;
     cors_allowed_origins?: string | string[];
   };
 } & Record<string, any>;

package/lib/index.mjs

@@ -6,7 +6,7 @@ from huggingface_hub import snapshot_download
 path = snapshot_download("${s}", revision="${o.model.revision||`main`}")
 print(path)
 `.trim(),a={...process.env};o.runtime.huggingface_token&&(a.HF_TOKEN=o.runtime.huggingface_token);let c=await q(t,[`-c`,n],{timeout:6e5,env:a});r?.(1);let l=c.stdout.trim().split(`
-`).pop();i?.({localPath:l,repoId:s,alreadyExists:!1})}catch(e){throw a?.(e),e}finally{t?.deleteDownload(l)}})();return t?.setDownload(l,u),{started:!0,localPath:null,repoId:s}}async function J(e,t=null,n={}){if(e===`ggml-llm`)return cn(t,n);if(e===`ggml-stt`)return Yn(t,n);if(e===`mlx-llm`)return kr(t,n);throw Error(`Unknown backend type: ${e}`)}var Y=`@fugood/buttress-backend-core`,X=`2.24.1`,jr={name:Y,private:!0,type:`module`,version:X,main:`src/index.js`,types:`lib/types/index.d.ts`,scripts:{build:`tsc --noCheck --declaration --emitDeclarationOnly --allowJs --outDir lib/types src/index.js`},dependencies:{"@fugood/buttress-hardware-guardrails":`^2.24.0`,"@fugood/llama.node":`^1.7.0-rc.11`,"@fugood/whisper.node":`^1.0.18`,"@huggingface/gguf":`^0.3.2`,"@iarna/toml":`^3.0.0`,bytes:`^3.1.0`}};const Mr=e=>{if(!e)return{repoId:null,filename:null};let[t,n]=e.split(`:`);return{repoId:t,filename:n||null}};async function Nr({modelIds:e=[],defaultConfig:t=null}={}){let n=[];console.log(`${Y} v${X}`),console.log(`Generating model capabilities comparison...
+`).pop();i?.({localPath:l,repoId:s,alreadyExists:!1})}catch(e){throw a?.(e),e}finally{t?.deleteDownload(l)}})();return t?.setDownload(l,u),{started:!0,localPath:null,repoId:s}}async function J(e,t=null,n={}){if(e===`ggml-llm`)return cn(t,n);if(e===`ggml-stt`)return Yn(t,n);if(e===`mlx-llm`)return kr(t,n);throw Error(`Unknown backend type: ${e}`)}var Y=`@fugood/buttress-backend-core`,X=`2.25.0-beta.5`,jr={name:Y,private:!0,type:`module`,version:X,main:`src/index.js`,types:`lib/types/index.d.ts`,scripts:{build:`tsc --noCheck --declaration --emitDeclarationOnly --allowJs --outDir lib/types src/index.js`},dependencies:{"@fugood/buttress-hardware-guardrails":`^2.25.0-beta.5`,"@fugood/llama.node":`^1.7.0-rc.11`,"@fugood/whisper.node":`^1.0.18`,"@huggingface/gguf":`^0.3.2`,"@iarna/toml":`^3.0.0`,bytes:`^3.1.0`}};const Mr=e=>{if(!e)return{repoId:null,filename:null};let[t,n]=e.split(`:`);return{repoId:t,filename:n||null}};async function Nr({modelIds:e=[],defaultConfig:t=null}={}){let n=[];console.log(`${Y} v${X}`),console.log(`Generating model capabilities comparison...
 `),n.push(`${Y} v${X}`),n.push(`## Model Capabilities Comparison
 `),(!e||e.length===0)&&(console.error(`Error: No model IDs provided`),process.exit(1));try{let r=(e={},t={})=>{let n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&typeof t==`object`&&!Array.isArray(t)?n[e]=r(n[e]||{},t):n[e]=t}),n},{server:i,generators:a=[],...o}=t||{},s=e=>r(structuredClone(o),e||{}),c=e=>{if(Array.isArray(a)&&a.length>0){let t=a.filter(e=>e?.type===`ggml-llm`);if(t.length>0&&e){let n=t.find(t=>t.model?.repo_id===e);if(n)return s(n)}}return Object.keys(o).length>0?s({}):null},l=[];for(let t=0;t<e.length;t+=1){let n=e[t];console.log(`[${t+1}/${e.length}] Analyzing ${n}...`);let r=c(n);r={...r||{},model:{...o.runtime,...r?.model||{},repo_id:n}};let i=await J(`ggml-llm`,null,{config:r,includeBreakdown:!0});l.push({modelId:n,capabilities:i,modelInfo:i.buttress?.selected||null,modelConfig:i.modelConfig||null})}let u=e=>e?(e/1024/1024/1024).toFixed(2):`N/A`,d=e=>e?`✅`:`🚫`;n.push(`| Model ID | Quantization | Size (GB) | Context Size | KV Cache Size (GB) | Total Required Memory (GB) | Fits GPU (Full) | Fits CPU (Full) |`),n.push(`|----------|--------------|-----------|--------------|--------------------|-----------------------------|-----------------|-----------------|`),l.forEach(({modelId:e,modelInfo:t,modelConfig:r})=>{let i=t?.quantization?.name?.toUpperCase()||`N/A`,a=u(t?.modelBytes),o=r?.nCtx||t?.kvInfo?.nCtxTrain||`N/A`,s=Ue(t),c=Number(o),l=t?.kvCacheBytes||(s&&Number.isFinite(c)&&c>0?s(c):s&&s(t?.kvInfo?.nCtxTrain||0))||null,f=u(l),p=u(t?.modelBytes&&l?t.modelBytes+l:t?.fit?.totalRequiredBytes),m=d(t?.fit?.fitsInGpu),h=d(t?.fit?.fitsInCpu);n.push(`| ${e} | ${i} | ${a} | ${o} | ${f} | ${p} | ${m} | ${h} |`);let g=t?.memoryLimitedCtx!=null||t?.limitedFit!=null,_=!t?.fit?.fitsInGpu||!t?.fit?.fitsInCpu;if(g&&_){let e=t?.memoryLimitedCtx||o,r=Number(e),i=t?.limitedKvCacheBytes||s&&Number.isFinite(r)&&r>0&&s(r)||null,c=u(i),l=u(t?.modelBytes&&i?t.modelBytes+i:t?.limitedFit?.totalRequiredBytes),m=d(t?.limitedFit?.fitsInGpu),h=d(t?.limitedFit?.fitsInCpu);(e!==o||c!==f||l!==p)&&n.push(`| ↳ Limited | - | ${a} | ${e} | ${c} | ${l} | ${m} | ${h} |`)}}),n.push(`
 ---`),n.push(`
@@ -33,7 +33,7 @@ print(path)
 === Hardware Information ===`);let t=null;if(process.platform!==`win32`)try{t=O(`uname -a`,{encoding:`utf8`}).trim()}catch{}t?console.log(`System: ${t}`):(console.log(`Hostname: ${v.hostname()}`),console.log(`OS: ${v.type()} ${v.release()}`)),console.log(`Platform: ${e.platform}`),console.log(`CPU Cores: ${v.cpus().length}`),console.log(`Total System Memory: ${(v.totalmem()/1024/1024/1024).toFixed(2)} GB`);let n=e.cpuTotalBytes>0?(e.cpuUsableBytes/e.cpuTotalBytes*100).toFixed(0):0;console.log(`Usable CPU Memory: ${(e.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),e.hasGpu?(console.log(`
 --- GPU Details ---`),e.devices.filter(e=>e.type===`gpu`).forEach(t=>{console.log(`GPU Backend: ${t.backend}`),console.log(`GPU Name: ${t.deviceName}`),console.log(`GPU Total Memory: ${(t.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let n=e.gpuTotalBytes>0?(e.gpuUsableBytes/e.gpuTotalBytes*100).toFixed(0):0;console.log(`GPU Usable Memory: ${(e.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.metadata&&(t.metadata.hasBFloat16&&console.log(`Supports BFloat16: Yes`),t.metadata.hasUnifiedMemory&&console.log(`Unified Memory: Yes`))})):console.log(`GPU: Not available`),console.log(`\nBackend Variant: ${e.variant}`),console.log(`Performance Score: ${e.score}`),e.fit&&(console.log(`
 --- Model Fit Analysis ---`),console.log(`Fits in GPU: ${e.fit.fitsInGpu?`Yes`:`No`}`),console.log(`Fits in CPU: ${e.fit.fitsInCpu?`Yes`:`No`}`),console.log(`Limiting Factor: ${e.fit.limiting}`))}console.log(`
-=== Full Capabilities JSON ===`),console.log(JSON.stringify(u,null,2)),process.exit(0)}catch(e){console.error(`Failed to get capabilities:`,e.message),process.exit(1)}}var Lr=e({finalizeGenerator:()=>Vr,generatorRegistry:()=>Z,getCapabilities:()=>J,getModelIdentifier:()=>Gr,ggmlLlm:()=>Hr,ggmlStt:()=>Wr,globalDownloadManager:()=>Rr,mlxLlm:()=>Ur,showModelsTable:()=>Nr,showSttModelsTable:()=>Fr,startGenerator:()=>Br,startModelDownload:()=>qr,status:()=>Kr,testGgmlLlmCapabilities:()=>Pr,testGgmlSttCapabilities:()=>Ir});const Z=new Map,Rr={downloads:new Map,getDownload(e){return this.downloads.get(e)||null},setDownload(e,t){this.downloads.set(e,t)},deleteDownload(e){this.downloads.delete(e)},isDownloading(e){return this.downloads.has(e)},getActiveDownloads(){return Array.from(this.downloads.entries()).map(([e,t])=>({localPath:e,promise:t}))}},zr=e=>{let t=Z.get(e);if(!t)throw Error(`Unknown generator id "${e}"`);return t},Q=(e,t)=>{let n=zr(e);if(n.type!==t)throw Error(`Generator "${e}" does not support ${t} backend`);return n.instance};async function Br(e,t){let n={"ggml-llm":{create:nn,getId:rn},"ggml-stt":{create:Gn,getId:Kn},"mlx-llm":{create:Dr,getId:Or}}[e];if(!n)throw Error(`Unsupported backend type: ${e}`);let r=n.getId(t);if(!r)throw Error(`Buttress generator config missing repo identifier`);let i=`${e}:${r}`,a=Z.get(i);if(a)return a.refCount+=1,a.instance.resetFinalized?.(),{id:a.id,info:a.instance.info};let o=await n.create(i,t,{globalDownloadManager:Rr}),s={id:i,type:o.type,instance:o,refCount:1};return Z.set(i,s),{id:i,info:o.info}}async function Vr(e){let t=Z.get(e);return t?(--t.refCount,t.refCount<=0&&(await t.instance.finalize(),(t.instance.hasPendingReleases?.()??!1)||Z.delete(e)),!0):!1}const Hr={async initContext(e,t){return Q(e,`ggml-llm`).initContext(t)},async completion(e,t){return Q(e,`ggml-llm`).completion(t)},async tokenize(e,t){return Q(e,`ggml-llm`).tokenize(t)},async detokenize(e,t){return Q(e,`ggml-llm`).detokenize(t)},async applyChatTemplate(e,t){return Q(e,`ggml-llm`).applyChatTemplate(t)},async releaseContext(e,t){let n=Z.get(e);if(!n)return{released:!0,alreadyReleased:!0};if(n.type!==`ggml-llm`)throw Error(`Generator "${e}" does not support ggml-llm backend`);return n.instance.releaseContext(t)}},Ur={async initContext(e,t){return Q(e,`mlx-llm`).initContext(t)},async completion(e,t){return Q(e,`mlx-llm`).completion(t)},async tokenize(e,t){return Q(e,`mlx-llm`).tokenize(t)},async detokenize(e,t){return Q(e,`mlx-llm`).detokenize(t)},async applyChatTemplate(e,t){return Q(e,`mlx-llm`).applyChatTemplate(t)},async releaseContext(e,t){let n=Z.get(e);if(!n)return{released:!0,alreadyReleased:!0};if(n.type!==`mlx-llm`)throw Error(`Generator "${e}" does not support mlx-llm backend`);return n.instance.releaseContext(t)}},Wr={async initContext(e,t){return Q(e,`ggml-stt`).initContext(t)},async transcribe(e,t){return Q(e,`ggml-stt`).transcribe(t)},async transcribeData(e,t){return Q(e,`ggml-stt`).transcribeData(t)},async releaseContext(e,t){let n=Z.get(e);if(!n)return{released:!0,alreadyReleased:!0};if(n.type!==`ggml-stt`)throw Error(`Generator "${e}" does not support ggml-stt backend`);return n.instance.releaseContext(t)}};function Gr(e,t){return e===`ggml-llm`?rn(t):e===`ggml-stt`?Kn(t):e===`mlx-llm`?Or(t):null}const Kr={getFullStatus:()=>et(Z),getGgmlLlmStatus:()=>Ze(Z),getGgmlSttStatus:()=>Qe(Z),getMlxLlmStatus:()=>$e(Z),subscribeToStatus:Ye,subscribeToStatusWithId:Xe,llmStatusTracker:H,sttStatusTracker:U,statusEmitter:V};async function qr(e,t,n={}){let r={"ggml-llm":an,"ggml-stt":qn,"mlx-llm":Ar}[e];return r?r(t,Rr,n):{started:!1,localPath:null,repoId:null,error:`Unknown backend type: ${e}`}}var Jr=`@fugood/buttress-server`,Yr=`2.24.1`,Xr={name:Jr,version:Yr,main:`lib/index.mjs`,types:`lib/index.d.mts`,type:`module`,bin:{"bricks-buttress":`./bin/bricks-buttress`},files:[`lib`,`bin`,`config`,`public`],scripts:{typecheck:`tsc --noEmit`,build:`tsdown -c rolldown.config.js`,prepublish:`bun run build`,dev:`bun src/index.ts`,start:`bun lib/index.mjs`,"start-with-node":`node lib/index.mjs`},keywords:[`BRICKS`,`buttress`,`server`],license:`MIT`,dependencies:{"@elysiajs/cors":`^1.1.1`,"@elysiajs/node":`^1.4.2`,"@fugood/llama.node":`^1.7.0-rc.11`,"@fugood/whisper.node":`^1.0.18`,"@huggingface/gguf":`^0.3.2`,"@iarna/toml":`^3.0.0`,bytes:`^3.1.0`,elysia:`^1.4.19`,ms:`^2.1.1`,"node-machine-id":`^1.1.12`,zod:`^3.25.76`},devDependencies:{tsdown:`^0.20.1`,typescript:`^5.9.3`},gitHead:`984a440ed04862f12c65f3cf62bdc70a938fcdd6`};const Zr=()=>({version:Yr,name:Jr,description:Xr.description}),Qr=typeof process<`u`&&process.versions&&process.versions.node,$r=e=>new n({adapter:Qr?t():void 0,...e}),ei=a.Object({id:a.String(),name:a.String(),version:a.String(),generators:a.Array(a.Object({type:a.String()})),authentication:a.Object({required:a.Boolean(),type:a.Literal(`device-group`)})}),ti=({store:{serverInfo:e}})=>({id:e.id,name:e.name,version:e.version,generators:e.generators,authentication:e.authentication});var ni=e=>{let t=$r(),n=e.autodiscover.http?.path??`/buttress/info`;return t.get(n,ti,{response:ei}),t};const ri=typeof process<`u`&&process.versions!=null&&process.versions.node!=null;var ii=$r().post(`/buttress/upload`,async({body:{file:e},store:{config:t}})=>{let n=`${Date.now()}-${e.name.replace(/[^\dA-Za-z]/g,`_`)}`,r=_.join(t.server.temp_file_dir,n);try{return ri?await g(r,await e.stream()):await g(r,await e.arrayBuffer()),{ok:!0,filename:n}}catch(e){return{ok:!1,error:String(e)}}},{body:a.Object({file:a.File()}),response:a.Object({ok:a.Boolean(),filename:a.Optional(a.String()),error:a.Optional(a.String())})}).get(`/buttress/download/:filename`,async({params:{filename:e},store:{config:t},status:n})=>{let i=_.join(t.server.temp_file_dir,e);return _.relative(t.server.temp_file_dir,i).includes(`..`)?(n(400),`Invalid file path`):r(i)},{params:a.Object({filename:a.String()})});const ai=_.dirname(D(import.meta.url)),oi=async()=>{let e=[_.join(ai,`..`,`public`,`status.html`),_.join(ai,`..`,`..`,`public`,`status.html`)];return(await Promise.all(e.map(e=>c.access(e).then(()=>e,()=>null)))).find(e=>e!==null)??null},si=e=>{let{status:t}=e;return t?.getFullStatus?t.getFullStatus():{timestamp:new Date().toISOString(),ggmlLlm:{generators:[],history:{}},ggmlStt:{generators:[],history:{}}}},ci=async()=>{let e=await oi();if(!e)return console.error(`[Status] Failed to find status.html in candidate paths`),new Response(`Status page not found`,{status:404,headers:{"Content-Type":`text/plain`}});try{let t=await c.readFile(e,`utf-8`);return new Response(t,{headers:{"Content-Type":`text/html; charset=utf-8`}})}catch(e){return console.error(`[Status] Failed to serve status page:`,e),new Response(`Status page not found`,{status:404,headers:{"Content-Type":`text/plain`}})}};var li=$r().get(`/status`,ci).get(`/status/`,ci).get(`/buttress/status`,({store:{backend:e}})=>si(e));const ui=[`ggml-llm`,`mlx-llm`],di=new Map;function fi(e,t){return t===`mlx-llm`?e.mlxLlm:e.ggmlLlm}async function pi(e,t,n,r=`[LLM]`){let i=(t.generators||[]).filter(e=>ui.includes(e.type));if(i.length===0)throw Error(`No LLM generator configured. Add a [[generators]] with type = "ggml-llm" or "mlx-llm" to your config.`);let a=i[0],o=n||a.model?.repo_id;if(n){let e=i.find(e=>e.model?.repo_id===n);e&&(a=e)}else o=a.model?.repo_id;let s=a.type||`ggml-llm`,c=o,l=di.get(c);if(l?.initialized)return l;let{generators:u,server:d,...f}=t.global||{},p={...f,...a,model:{...a.model,repo_id:o}};console.log(`${r} Creating ${s} generator for ${c}`);let{id:m}=await e.startGenerator(s,p),h={id:m,type:s,config:p,repoId:o,initialized:!1};return di.set(c,h),await fi(e,s).initContext(m,{}),h.initialized=!0,console.log(`${r} Generator ready: ${c}`),h}function mi(e){let t=e.timings||{},n=t.prompt_n??t.promptN??0,r=t.cache_n??t.cacheN??0,i=t.predicted_n??t.predictedN??0;return{promptTokens:n||e.prompt_tokens||e.promptTokens||0,cachedTokens:r,completionTokens:i||e.tokens_evaluated||e.tokensEvaluated||e.tokens_predicted||e.tokensPredicted||0}}function hi(e){let{promptTokens:t,cachedTokens:n,completionTokens:r}=mi(e),i=t+n;return{prompt_tokens:i,completion_tokens:r,total_tokens:i+r}}const gi=()=>`chatcmpl-${Date.now()}-${Math.random().toString(36).slice(2,9)}`;async function _i(e,t,n,r){let i=e.getReader(),a=``,o=null,s=null,c=`stop`,l={prompt_tokens:0,completion_tokens:0,total_tokens:0};try{let e=!1;for(;!e;){let n=await i.read();if({done:e}=n,e)break;let{event:r,data:u}=n.value;if(r===`token`)u.content!=null&&(a=u.content),u.reasoning_content!=null&&(o=u.reasoning_content);else if(r===`result`)u.content==null?u.text&&(a=u.text):a=u.content,u.reasoning_content!=null&&(o=u.reasoning_content),u.tool_calls?.length>0?(s=u.tool_calls.map((e,n)=>({id:e.id||`call_${t}_${n}`,type:`function`,function:{name:e.function?.name||``,arguments:e.function?.arguments||``}})),c=`tool_calls`):c=u.interrupted?`length`:`stop`,l=hi(u);else if(r===`error`)throw Error(u.message)}}finally{i.cancel().catch(()=>{})}let u={role:`assistant`,content:a||null};return o&&(u.reasoning_content=o),s&&(u.tool_calls=s),{id:t,object:`chat.completion`,created:n,model:r,choices:[{index:0,message:u,finish_reason:c}],usage:l}}function vi({global:e}){let t=$r({prefix:`/oai-compat`});return t.use(M({origin:e?.openai_compat?.cors_allowed_origins??!1,methods:[`GET`,`POST`,`OPTIONS`],allowedHeaders:[`Content-Type`,`Authorization`],maxAge:86400,preflight:!0})),t.get(`/v1/models`,({store:e})=>{let{config:t}=e,n=(t.generators||[]).filter(e=>ui.includes(e.type)).map(e=>({id:e.model?.repo_id||e.type,object:`model`,created:Math.floor(Date.now()/1e3),owned_by:`local`}));return n.length===0&&n.push({id:`ggml-llm`,object:`model`,created:Math.floor(Date.now()/1e3),owned_by:`local`}),{object:`list`,data:n}}),t.post(`/v1/chat/completions`,async function*({body:e,set:t,store:n}){let{config:r,backend:a}=n,{messages:o=[],stream:s=!1,model:c,tools:l,temperature:u,stop:d,top_p:f,max_tokens:p,presence_penalty:m,frequency_penalty:h,tool_choice:g,stream_options:_,enable_thinking:v}=e;if(!o||o.length===0)return t.status=400,{error:{message:`messages is required and must not be empty`,type:`invalid_request_error`}};try{let e=await pi(a,r,c,`[OpenAI]`),t=gi(),n=Math.floor(Date.now()/1e3),y=e.repoId||`ggml-llm`,b={reasoning_format:`auto`,messages:o,jinja:!0,add_generation_prompt:!0};u!=null&&(b.temperature=u),f!=null&&(b.top_p=f),p!=null&&(b.n_predict=p),d!=null&&(b.stop=Array.isArray(d)?d:[d]),m!=null&&(b.presence_penalty=m),h!=null&&(b.frequency_penalty=h),l!=null&&(b.tools=l),g!=null&&(b.tool_choice=g),b.enable_thinking=v??!1;let x=await fi(a,e.type).completion(e.id,{options:b});if(!s)return await _i(x,t,n,y);let S=_?.include_usage===!0,C=x.getReader(),w=``,T=``,E=new Map,D=new Map;try{let e=!1;for(;!e;){let r=await C.read();if({done:e}=r,e)break;let{event:a,data:o}=r.value;if(a===`token`){let e={};if(o.content!=null){let t=o.content;t.length>w.length&&(e.content=t.slice(w.length),w=t)}if(o.reasoning_content!=null){let t=o.reasoning_content;t.length>T.length&&(e.reasoning_content=t.slice(T.length),T=t)}if(o.tool_calls?.length>0){let n=[];o.tool_calls.forEach((e,r)=>{let i={index:r};D.has(r)||(D.set(r,e.id||`call_${t}_${r}`),i.id=D.get(r),i.type=`function`);let a=e.function?.arguments||``,o=E.get(r)||``,s={};!E.has(r)&&e.function?.name&&(s.name=e.function.name),a.length>o.length&&(s.arguments=a.slice(o.length),E.set(r,a)),Object.keys(s).length>0?(i.function=s,n.push(i)):i.id&&(i.function={name:e.function?.name||``,arguments:``},n.push(i))}),n.length>0&&(e.tool_calls=n)}Object.keys(e).length>0&&(yield i({data:JSON.stringify({id:t,object:`chat.completion.chunk`,created:n,model:y,choices:[{index:0,delta:e,finish_reason:null}]})}))}else if(a===`result`){let e=`stop`;o.tool_calls?.length>0||D.size>0?e=`tool_calls`:o.interrupted&&(e=`length`);let r={id:t,object:`chat.completion.chunk`,created:n,model:y,choices:[{index:0,delta:{},finish_reason:e}]};S&&(r.usage=hi(o)),yield i({data:JSON.stringify(r)})}else a===`error`&&(yield i({data:JSON.stringify({error:{message:o.message,type:`server_error`}})}))}yield i({data:`[DONE]`})}finally{C.cancel().catch(()=>{})}}catch(e){return console.error(`[OpenAI] Chat completion error:`,e),t.status=500,{error:{message:e.message||`Internal server error`,type:`server_error`}}}},{body:a.Object({model:a.Optional(a.String()),messages:a.Array(a.Any()),stream:a.Optional(a.Boolean()),temperature:a.Optional(a.Number()),top_p:a.Optional(a.Number()),max_tokens:a.Optional(a.Number()),stop:a.Optional(a.Union([a.String(),a.Array(a.String())])),presence_penalty:a.Optional(a.Number()),frequency_penalty:a.Optional(a.Number()),tools:a.Optional(a.Array(a.Any())),tool_choice:a.Optional(a.Any()),stream_options:a.Optional(a.Object({include_usage:a.Optional(a.Boolean())})),enable_thinking:a.Optional(a.Boolean())})}),t}const yi=()=>`msg_${Date.now()}${Math.random().toString(36).slice(2,11)}`;function bi(e){let t={},n=[];if(e.system!=null){let t=``;if(typeof e.system==`string`)t=e.system;else if(Array.isArray(e.system))for(let n of e.system)n?.type===`text`&&typeof n.text==`string`&&(t+=n.text);t&&n.push({role:`system`,content:t})}if(!Array.isArray(e.messages))throw Error(`'messages' is required and must be an array`);for(let t of e.messages){let e=t?.role||`user`;if(t?.content==null){if(e===`assistant`)continue;n.push(t);continue}if(typeof t.content==`string`){n.push({role:e,content:t.content});continue}if(!Array.isArray(t.content)){n.push(t);continue}let r=[],i=[],a=[],o=``,s=!1;for(let e of t.content){let t=e?.type||``;if(t===`text`)i.push({type:`text`,text:e.text||``});else if(t===`thinking`)o+=e.thinking||``;else if(t===`image`){let t=e.source||{};if(t.type===`base64`){let e=t.media_type||`image/jpeg`,n=t.data||``;i.push({type:`image_url`,image_url:{url:`data:${e};base64,${n}`}})}else t.type===`url`&&i.push({type:`image_url`,image_url:{url:t.url||``}})}else if(t===`tool_use`)r.push({id:e.id||``,type:`function`,function:{name:e.name||``,arguments:JSON.stringify(e.input??{})}}),s=!0;else if(t===`tool_result`){let t=e.tool_use_id||``,n=``,r=e.content;if(typeof r==`string`)n=r;else if(Array.isArray(r))for(let e of r)e?.type===`text`&&(n+=e.text||``);a.push({role:`tool`,tool_call_id:t,content:n})}}if(i.length>0||s||o){let t={role:e};i.length>0?t.content=i:(s||o)&&(t.content=``),r.length>0&&(t.tool_calls=r),o&&(t.reasoning_content=o),n.push(t)}for(let e of a)n.push(e)}if(t.messages=n,Array.isArray(e.tools)&&(t.tools=e.tools.map(e=>({type:`function`,function:{name:e.name||``,description:e.description||``,parameters:e.input_schema||{}}}))),e.tool_choice&&typeof e.tool_choice==`object`){let n=e.tool_choice.type;n===`auto`?t.tool_choice=`auto`:n===`any`||n===`tool`?t.tool_choice=`required`:n===`none`&&(t.tool_choice=`none`)}else Array.isArray(t.tools)&&t.tools.length>0&&(t.tool_choice=`auto`);e.stop_sequences!=null&&(t.stop=Array.isArray(e.stop_sequences)?e.stop_sequences:[e.stop_sequences]),t.max_tokens=e.max_tokens??4096;for(let n of[`temperature`,`top_p`,`top_k`,`stream`])e[n]!=null&&(t[n]=e[n]);return e.thinking&&typeof e.thinking==`object`&&e.thinking.type===`enabled`&&(t.enable_thinking=!0,e.thinking.budget_tokens!=null&&(t.thinking_budget_tokens=e.thinking.budget_tokens)),t}function xi(e,t){return t?`tool_use`:e.stopping_word||e.stoppingWord?`stop_sequence`:e.interrupted||e.truncated?`max_tokens`:`end_turn`}function Si(e){let{promptTokens:t,cachedTokens:n,completionTokens:r}=mi(e);return{cache_read_input_tokens:n,input_tokens:t,output_tokens:r}}async function Ci(e,t,n){let r=e.getReader(),i=``,a=``,o=[],s={cache_read_input_tokens:0,input_tokens:0,output_tokens:0},c=`end_turn`,l=null;try{let e=!1;for(;!e;){let t=await r.read();if({done:e}=t,e)break;let{event:n,data:u}=t.value;if(n===`token`)u.content!=null&&(i=u.content),u.reasoning_content!=null&&(a=u.reasoning_content);else if(n===`result`)u.content==null?u.text&&u.reasoning_content==null&&(i=u.text):i=u.content,u.reasoning_content!=null&&(a=u.reasoning_content),Array.isArray(u.tool_calls)&&(o=u.tool_calls),s=Si(u),l=u.stopping_word||u.stoppingWord||null,c=xi(u,o.length>0);else if(n===`error`)throw Error(u.message||`completion error`)}}finally{r.cancel().catch(()=>{})}let u=[];a&&u.push({type:`thinking`,thinking:a,signature:``}),i&&u.push({type:`text`,text:i});for(let e of o){let t={};try{t=JSON.parse(e.function?.arguments||`{}`)}catch{t={}}u.push({type:`tool_use`,id:e.id||`toolu_${Math.random().toString(36).slice(2,11)}`,name:e.function?.name||``,input:t})}return{id:t,type:`message`,role:`assistant`,content:u,model:n,stop_reason:c,stop_sequence:l,usage:s}}function wi({global:e}){let t=$r({prefix:`/anthropic-messages`});return t.use(M({origin:e?.anthropic_messages?.cors_allowed_origins??!1,methods:[`GET`,`POST`,`OPTIONS`],allowedHeaders:[`Content-Type`,`Authorization`,`x-api-key`,`anthropic-version`],maxAge:86400,preflight:!0})),t.post(`/v1/messages`,async function*({body:e,set:t,store:n}){let{config:r,backend:a}=n,o=e;try{if(!Array.isArray(o.messages)||o.messages.length===0)return t.status=400,{type:`error`,error:{type:`invalid_request_error`,message:`messages is required and must not be empty`}};let e=bi(o),n=await pi(a,r,o.model,`[Anthropic]`),s=yi(),c=n.repoId||`ggml-llm`,l={reasoning_format:`auto`,messages:e.messages,jinja:!0,add_generation_prompt:!0};e.temperature!=null&&(l.temperature=e.temperature),e.top_p!=null&&(l.top_p=e.top_p),e.top_k!=null&&(l.top_k=e.top_k),e.max_tokens!=null&&(l.n_predict=e.max_tokens),e.stop!=null&&(l.stop=e.stop),e.tools!=null&&(l.tools=e.tools),e.tool_choice!=null&&(l.tool_choice=e.tool_choice),l.enable_thinking=e.enable_thinking??!1,e.thinking_budget_tokens!=null&&(l.thinking_budget_tokens=e.thinking_budget_tokens);let u=await fi(a,n.type).completion(n.id,{options:l});if(!o.stream)return await Ci(u,s,c);let d=u.getReader(),f=``,p=``,m=new Map,h=new Map,g=new Map,_=new Set,v=!1,y=!1,b=0,x=0,S={cache_read_input_tokens:0,input_tokens:0,output_tokens:0},C=`end_turn`,w=null,T=!1,E=e=>(v?1:0)+(y?1:0)+e;try{let e=!1;for(;!e;){let t=await d.read();if({done:e}=t,e)break;let{event:n,data:r}=t.value;if(n===`token`){if(!T){let e=Si(r);yield i({event:`message_start`,data:JSON.stringify({type:`message_start`,message:{id:s,type:`message`,role:`assistant`,content:[],model:c,stop_reason:null,stop_sequence:null,usage:e}})}),T=!0}if(r.reasoning_content!=null){let e=r.reasoning_content;e.length>p.length&&(v||(yield i({event:`content_block_start`,data:JSON.stringify({type:`content_block_start`,index:0,content_block:{type:`thinking`,thinking:``}})}),v=!0,b=1),yield i({event:`content_block_delta`,data:JSON.stringify({type:`content_block_delta`,index:0,delta:{type:`thinking_delta`,thinking:e.slice(p.length)}})}),p=e)}if(r.content!=null){let e=r.content;e.length>f.length&&(y||=(yield i({event:`content_block_start`,data:JSON.stringify({type:`content_block_start`,index:b,content_block:{type:`text`,text:``}})}),!0),yield i({event:`content_block_delta`,data:JSON.stringify({type:`content_block_delta`,index:b,delta:{type:`text_delta`,text:e.slice(f.length)}})}),f=e)}if(Array.isArray(r.tool_calls)&&r.tool_calls.length>0){for(let e=0;e<r.tool_calls.length;e+=1){let t=r.tool_calls[e],n=E(e),a=t?.function?.arguments||``,o=m.get(e)||``;if(!_.has(e)){let r=t?.id||`toolu_${s}_${e}`,a=t?.function?.name||g.get(e)||``;h.set(e,r),g.set(e,a),_.add(e),yield i({event:`content_block_start`,data:JSON.stringify({type:`content_block_start`,index:n,content_block:{type:`tool_use`,id:r,name:a,input:{}}})})}a.length>o.length&&(yield i({event:`content_block_delta`,data:JSON.stringify({type:`content_block_delta`,index:n,delta:{type:`input_json_delta`,partial_json:a.slice(o.length)}})}),m.set(e,a))}x=r.tool_calls.length}}else if(n===`result`){if(!T){let e=Si(r);yield i({event:`message_start`,data:JSON.stringify({type:`message_start`,message:{id:s,type:`message`,role:`assistant`,content:[],model:c,stop_reason:null,stop_sequence:null,usage:e}})}),T=!0}Array.isArray(r.tool_calls)&&(x=Math.max(x,r.tool_calls.length)),S=Si(r),w=r.stopping_word||r.stoppingWord||null,C=xi(r,_.size>0)}else if(n===`error`){yield i({event:`error`,data:JSON.stringify({type:`error`,error:{type:`api_error`,message:r.message||`completion error`}})});return}}v&&(yield i({event:`content_block_delta`,data:JSON.stringify({type:`content_block_delta`,index:0,delta:{type:`signature_delta`,signature:``}})}),yield i({event:`content_block_stop`,data:JSON.stringify({type:`content_block_stop`,index:0})})),y&&(yield i({event:`content_block_stop`,data:JSON.stringify({type:`content_block_stop`,index:b})}));for(let e of[..._].sort((e,t)=>e-t))yield i({event:`content_block_stop`,data:JSON.stringify({type:`content_block_stop`,index:E(e)})});yield i({event:`message_delta`,data:JSON.stringify({type:`message_delta`,delta:{stop_reason:C,stop_sequence:w},usage:{output_tokens:S.output_tokens}})}),yield i({event:`message_stop`,data:JSON.stringify({type:`message_stop`})})}finally{d.cancel().catch(()=>{})}}catch(e){return console.error(`[Anthropic] Messages error:`,e),t.status=500,{type:`error`,error:{type:`api_error`,message:e?.message||`Internal server error`}}}},{body:a.Object({model:a.Optional(a.String()),messages:a.Array(a.Any()),system:a.Optional(a.Any()),max_tokens:a.Optional(a.Number()),stream:a.Optional(a.Boolean()),temperature:a.Optional(a.Number()),top_p:a.Optional(a.Number()),top_k:a.Optional(a.Number()),stop_sequences:a.Optional(a.Union([a.String(),a.Array(a.String())])),tools:a.Optional(a.Array(a.Any())),tool_choice:a.Optional(a.Any()),thinking:a.Optional(a.Any()),metadata:a.Optional(a.Any())})}),t.post(`/v1/messages/count_tokens`,async({body:e,set:t,store:n})=>{let{config:r,backend:i}=n,a=e;try{let e=bi(a),t=await pi(i,r,a.model,`[Anthropic]`),n=fi(i,t.type),o={messages:e.messages,add_generation_prompt:!0,jinja:!0};e.tools!=null&&(o.tools=e.tools);let s=await n.applyChatTemplate(t.id,o),c=typeof s==`string`?s:s?.prompt||``,l=await n.tokenize(t.id,{text:c,add_special:!0,parse_special:!0});return{input_tokens:(Array.isArray(l)?l:l?.tokens||[]).length}}catch(e){return console.error(`[Anthropic] count_tokens error:`,e),t.status=500,{type:`error`,error:{type:`api_error`,message:e?.message||`Internal server error`}}}},{body:a.Object({model:a.Optional(a.String()),messages:a.Array(a.Any()),system:a.Optional(a.Any()),tools:a.Optional(a.Array(a.Any())),tool_choice:a.Optional(a.Any())})}),t.get(`/v1/models`,({store:e})=>{let{config:t}=e,n=(t.generators||[]).filter(e=>ui.includes(e.type)).map(e=>{let t=e.model?.repo_id||e.type;return{id:t,type:`model`,display_name:t,created_at:new Date().toISOString()}});return n.length===0&&n.push({id:`ggml-llm`,type:`model`,display_name:`ggml-llm`,created_at:new Date().toISOString()}),{data:n,has_more:!1,first_id:n[0]?.id,last_id:n.at(-1)?.id}}),t}const Ti=(e={},t={})=>{let n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&typeof t==`object`&&!Array.isArray(t)?n[e]=Ti(n[e]||{},t):n[e]=t}),n},Ei=e=>e&&typeof e==`object`?structuredClone(e):null,Di=(e,t)=>Ti(Ei(e)||{},Ei(t)||{}),Oi=(e,t)=>Ti(structuredClone(e.global),t||{}),ki=(e,t,n,r)=>{if(e.generators.length>0){let i=e.generators.filter(e=>e?.type===n);if(i.length>0&&r){let a=i.find(e=>t.getModelIdentifier(n,e)===r);if(a)return Oi(e,a)}}return Object.keys(e.global).length>0?Oi(e,{}):null},Ai={udp:{port:8089,announcements:{enabled:!0,interval:5e3},requests:{enabled:!0,responseDelay:100}},http:{enabled:!0,path:`/buttress/info`,cors:!0}},ji=e=>e?e===!0?{...Ai}:Ti(Ai,e):null,Mi=(e,t)=>{if(!e.generators||e.generators.length===0)return t.map(e=>({type:e}));let n=new Set;return e.generators.forEach(e=>{e.type&&n.add(e.type)}),n.size===0?t.map(e=>({type:e})):Array.from(n).map(e=>({type:e}))},Ni=(e,t,n)=>e===void 0?n:typeof e==`number`?e:t(e)??n,Pi=6e4,Fi=1024*1024*50,Ii=e=>{let t=N.machineIdSync(),n=Ti({server:{id:`buttress-${t}`,name:`Buttress Server (${t.slice(-8)})`,port:2080,temp_file_dir:_.join(v.tmpdir(),`.buttress`),session_timeout:Pi,max_body_size:Fi},autodiscover:!1},Ei(e)||{}),r=Array.isArray(n.generators)?n.generators:[],{server:i,generators:a,autodiscover:o,...s}=n;return{autodiscover:ji(o),server:{id:i.id,name:i.name,port:i.port,log_level:i.log_level,temp_file_dir:i.temp_file_dir,max_body_size:Ni(i.max_body_size,w.parse,Fi),session_timeout:Ni(i.session_timeout,P,Pi)},global:s,generators:r}},Li={getCapabilities:j.tuple([j.object({type:j.string().optional().default(`ggml-llm`),config:j.any().optional(),currentClientCapabilities:j.any().optional(),options:j.any().optional()}).nullable().optional()]),startGenerator:j.tuple([j.string(),j.any().optional()]),finalizeGenerator:j.tuple([j.string()])};var Ri={async getCapabilities({backend:e,config:t},n=null){console.log(`[Server] Get Capabilities:`,n);let{type:r=`ggml-llm`,config:i,currentClientCapabilities:a=null,options:o={}}=n||{type:`ggml-llm`},s=Ei(i),c=Di(ki(t,e,r,e.getModelIdentifier(r,s)),i);if(Object.keys(c).length===0)throw Error(`Buttress server missing generator configuration`);return c.backend=c.backend||{},c.backend.type||(c.backend.type=r),e.getCapabilities(r,a,{...o,config:c})},async startGenerator({backend:e,config:t,session:n},r,i){console.log(`[Server] Start Generator:`,r,i);let a=Ei(i),o=Di(ki(t,e,r,e.getModelIdentifier(r,a)),i);if(Object.keys(o).length===0)throw Error(`Buttress server missing generator configuration`);o.backend=o.backend||{},o.backend.type||(o.backend.type=r);let s=await e.startGenerator(r,o);return n.generators.add(s.id),s},async finalizeGenerator({backend:e,session:t},n){return console.log(`[Server] Finalize Generator:`,n),t.generators.delete(n),e.finalizeGenerator(n)}};const zi={initContext:j.tuple([j.string(),j.any().optional()]),completion:j.tuple([j.string(),j.any().optional()]),tokenize:j.tuple([j.string(),j.any()]),detokenize:j.tuple([j.string(),j.any()]),applyChatTemplate:j.tuple([j.string(),j.any()]),releaseContext:j.tuple([j.string()])};function Bi(e){return function({backend:t,session:n},r,i){return new s({async start(a){try{let o=await e(t).initContext(r,{...i,onProgress:e=>{a.enqueue({event:`progress`,data:{progress:e}})}});n.initializedContexts.add(r),await new Promise(e=>setTimeout(e,1e3));let{download:s,...c}=o||{};a.enqueue({event:`result`,data:{result:c}}),a.close()}catch(e){a.error(e)}}})}}function Vi(e,t){return async function({backend:n,session:r},i,a){return console.log(`[Server] ${t}:`,{id:i,force:a}),r.initializedContexts.has(i)?(r.initializedContexts.delete(i),e(n).releaseContext(i,{force:a})):(console.log(`[Server] ${t} skipped - not initialized by this session:`,{id:i}),{released:!1,skipped:!0})}}function Hi(e,t){return{initContext:Bi(e),completion({backend:n},r,i){return console.log(`[Server] ${t}Completion:`,{id:r,property:i}),e(n).completion(r,i)},async tokenize({backend:n},r,i){return console.log(`[Server] ${t}Tokenize:`,{id:r,property:i}),e(n).tokenize(r,i)},async detokenize({backend:n},r,i){return console.log(`[Server] ${t}Detokenize:`,{id:r,property:i}),e(n).detokenize(r,i)},async applyChatTemplate({backend:n},r,i){return console.log(`[Server] ${t}Apply Chat Template:`,{id:r,property:i}),e(n).applyChatTemplate(r,i)},releaseContext:Vi(e,`${t}Release Context`)}}var Ui=Hi(e=>e.ggmlLlm,``);const Wi={initContext:j.tuple([j.string(),j.any().optional()]),transcribe:j.tuple([j.string(),j.string(),j.any().optional()]),transcribeData:j.tuple([j.string(),j.union([j.instanceof(Buffer),j.instanceof(Uint8Array)]),j.any().optional()]),releaseContext:j.tuple([j.string()])},Gi=e=>e.ggmlStt,Ki={common:Ri,ggmlLlm:Ui,ggmlStt:{initContext:Bi(Gi),async transcribe({backend:e,config:{server:t}},n,r,i){return console.log(`[Server] Transcribe:`,{id:n,audioPath:r,options:i}),e.ggmlStt.transcribe(n,{audioPath:_.join(t.temp_file_dir,r),options:i})},async transcribeData({backend:e},t,n,r){return console.log(`[Server] Transcribe Data:`,{id:t,audioDataLength:n?.length||0,options:r}),e.ggmlStt.transcribeData(t,{audioData:n,options:r})},releaseContext:Vi(Gi,`Release STT Context`)},mlxLlm:Hi(e=>e.mlxLlm,`MLX `)},qi={common:Li,ggmlLlm:zi,ggmlStt:Wi,mlxLlm:zi};var Ji=Ki;const Yi=e=>{try{return JSON.parse(e,(e,t)=>{if(!t)return t;if(t?.type===`Buffer`&&t?.data)return F.from(t.data,`base64`);if(t?.type===`Uint8Array`&&t?.data){let e=F.from(t.data,`base64`);return e.buffer.slice(e.byteOffset,e.byteOffset+e.byteLength)}return t?.type===`Error`&&t?.name&&t?.message?Error(t.name,t.message):t})}catch{return e}},Xi=e=>{try{return JSON.stringify(e,(e,t)=>t instanceof Error?{type:`Error`,name:t.name,message:t.message}:t instanceof F?{type:`Buffer`,data:t.toString(`base64`)}:t instanceof Uint8Array?{type:`Uint8Array`,data:F.from(t).toString(`base64`)}:t)}catch{return e}};var Zi=class{name=`udp`;socket=null;announcementTimer=null;config;getServerInfo;port;constructor(e,t){this.config=e,this.getServerInfo=t,this.port=e.port??8089}async start(){if(this.socket=re.createSocket({type:`udp4`,reuseAddr:!0}),this.socket.on(`message`,(e,t)=>{this.handleMessage(e,t)}),this.socket.on(`error`,e=>{console.error(`[Autodiscover UDP] Socket error:`,e.message)}),await new Promise((e,t)=>{this.socket.bind(this.port,()=>{this.socket.setBroadcast(!0),console.log(`[Autodiscover UDP] Listening on port ${this.port}`),e()}),this.socket.once(`error`,t)}),this.config.announcements.enabled){let e=this.config.announcements.interval??5e3;this.announcementTimer=setInterval(()=>{this.sendAnnouncement()},e),this.sendAnnouncement()}}async stop(){this.announcementTimer&&=(clearInterval(this.announcementTimer),null),this.socket&&=(await new Promise(e=>{this.socket.close(()=>e())}),null)}handleMessage(e,t){try{let n=JSON.parse(e.toString());if(n.t===`QUERY`&&this.config.requests.enabled){let e=n.d,r=this.config.requests.responseDelay??0,i=r>0?Math.random()*r:0;setTimeout(()=>{this.sendResponse(e.id,t)},i)}}catch{}}sendAnnouncement(){if(!this.socket)return;let e={t:`ANNOUNCE`,v:`1.0`,d:{info:this.getServerInfo()}},t=Buffer.from(JSON.stringify(e));this.socket.send(t,0,t.length,this.port,`255.255.255.255`,e=>{e&&console.error(`[Autodiscover UDP] Announcement error:`,e.message)})}sendResponse(e,t){if(!this.socket)return;let n={t:`RESPONSE`,v:`1.0`,d:{request_id:e,info:this.getServerInfo()}},r=Buffer.from(JSON.stringify(n));this.socket.send(r,0,r.length,t.port,t.address,e=>{e&&console.error(`[Autodiscover UDP] Response error:`,e.message)})}},Qi=class{transports=[];started=!1;constructor(e,t){this.config=e,this.getServerInfo=t,(e.udp?.announcements?.enabled||e.udp?.requests?.enabled)&&this.transports.push(new Zi(e.udp,t))}async start(){this.started||=((await Promise.allSettled(this.transports.map(e=>e.start()))).forEach((e,t)=>{e.status===`rejected`&&console.error(`[Autodiscover] Failed to start ${this.transports[t].name}:`,e.reason)}),!0)}async stop(){this.started&&=(await Promise.allSettled(this.transports.map(e=>e.stop())),!1)}};const $i=()=>{let e=v.networkInterfaces();return Object.values(e).flat().find(e=>e?.family===`IPv4`&&!e?.internal)?.address||null},$=Zr(),ea=e=>{if(!e)return{repoId:null,filename:null};let[t,n]=e.split(`:`);return{repoId:t,filename:n||null}};async function ta({modelIds:e=[],defaultConfig:t=null}={}){let n=[];console.log(`${$.name} v${$.version}`),console.log(`Generating model capabilities comparison...
+=== Full Capabilities JSON ===`),console.log(JSON.stringify(u,null,2)),process.exit(0)}catch(e){console.error(`Failed to get capabilities:`,e.message),process.exit(1)}}var Lr=e({finalizeGenerator:()=>Vr,generatorRegistry:()=>Z,getCapabilities:()=>J,getModelIdentifier:()=>Gr,ggmlLlm:()=>Hr,ggmlStt:()=>Wr,globalDownloadManager:()=>Rr,mlxLlm:()=>Ur,showModelsTable:()=>Nr,showSttModelsTable:()=>Fr,startGenerator:()=>Br,startModelDownload:()=>qr,status:()=>Kr,testGgmlLlmCapabilities:()=>Pr,testGgmlSttCapabilities:()=>Ir});const Z=new Map,Rr={downloads:new Map,getDownload(e){return this.downloads.get(e)||null},setDownload(e,t){this.downloads.set(e,t)},deleteDownload(e){this.downloads.delete(e)},isDownloading(e){return this.downloads.has(e)},getActiveDownloads(){return Array.from(this.downloads.entries()).map(([e,t])=>({localPath:e,promise:t}))}},zr=e=>{let t=Z.get(e);if(!t)throw Error(`Unknown generator id "${e}"`);return t},Q=(e,t)=>{let n=zr(e);if(n.type!==t)throw Error(`Generator "${e}" does not support ${t} backend`);return n.instance};async function Br(e,t){let n={"ggml-llm":{create:nn,getId:rn},"ggml-stt":{create:Gn,getId:Kn},"mlx-llm":{create:Dr,getId:Or}}[e];if(!n)throw Error(`Unsupported backend type: ${e}`);let r=n.getId(t);if(!r)throw Error(`Buttress generator config missing repo identifier`);let i=`${e}:${r}`,a=Z.get(i);if(a)return a.refCount+=1,a.instance.resetFinalized?.(),{id:a.id,info:a.instance.info};let o=await n.create(i,t,{globalDownloadManager:Rr}),s={id:i,type:o.type,instance:o,refCount:1};return Z.set(i,s),{id:i,info:o.info}}async function Vr(e){let t=Z.get(e);return t?(--t.refCount,t.refCount<=0&&(await t.instance.finalize(),(t.instance.hasPendingReleases?.()??!1)||Z.delete(e)),!0):!1}const Hr={async initContext(e,t){return Q(e,`ggml-llm`).initContext(t)},async completion(e,t){return Q(e,`ggml-llm`).completion(t)},async tokenize(e,t){return Q(e,`ggml-llm`).tokenize(t)},async detokenize(e,t){return Q(e,`ggml-llm`).detokenize(t)},async applyChatTemplate(e,t){return Q(e,`ggml-llm`).applyChatTemplate(t)},async releaseContext(e,t){let n=Z.get(e);if(!n)return{released:!0,alreadyReleased:!0};if(n.type!==`ggml-llm`)throw Error(`Generator "${e}" does not support ggml-llm backend`);return n.instance.releaseContext(t)}},Ur={async initContext(e,t){return Q(e,`mlx-llm`).initContext(t)},async completion(e,t){return Q(e,`mlx-llm`).completion(t)},async tokenize(e,t){return Q(e,`mlx-llm`).tokenize(t)},async detokenize(e,t){return Q(e,`mlx-llm`).detokenize(t)},async applyChatTemplate(e,t){return Q(e,`mlx-llm`).applyChatTemplate(t)},async releaseContext(e,t){let n=Z.get(e);if(!n)return{released:!0,alreadyReleased:!0};if(n.type!==`mlx-llm`)throw Error(`Generator "${e}" does not support mlx-llm backend`);return n.instance.releaseContext(t)}},Wr={async initContext(e,t){return Q(e,`ggml-stt`).initContext(t)},async transcribe(e,t){return Q(e,`ggml-stt`).transcribe(t)},async transcribeData(e,t){return Q(e,`ggml-stt`).transcribeData(t)},async releaseContext(e,t){let n=Z.get(e);if(!n)return{released:!0,alreadyReleased:!0};if(n.type!==`ggml-stt`)throw Error(`Generator "${e}" does not support ggml-stt backend`);return n.instance.releaseContext(t)}};function Gr(e,t){return e===`ggml-llm`?rn(t):e===`ggml-stt`?Kn(t):e===`mlx-llm`?Or(t):null}const Kr={getFullStatus:()=>et(Z),getGgmlLlmStatus:()=>Ze(Z),getGgmlSttStatus:()=>Qe(Z),getMlxLlmStatus:()=>$e(Z),subscribeToStatus:Ye,subscribeToStatusWithId:Xe,llmStatusTracker:H,sttStatusTracker:U,statusEmitter:V};async function qr(e,t,n={}){let r={"ggml-llm":an,"ggml-stt":qn,"mlx-llm":Ar}[e];return r?r(t,Rr,n):{started:!1,localPath:null,repoId:null,error:`Unknown backend type: ${e}`}}var Jr=`@fugood/buttress-server`,Yr=`2.25.0-beta.6`,Xr={name:Jr,version:Yr,main:`lib/index.mjs`,types:`lib/index.d.mts`,type:`module`,bin:{"bricks-buttress":`./bin/bricks-buttress`},files:[`lib`,`bin`,`config`,`public`],scripts:{typecheck:`tsc --noEmit`,build:`tsdown -c rolldown.config.js`,prepublish:`bun run build`,dev:`bun src/index.ts`,start:`bun lib/index.mjs`,"start-with-node":`node lib/index.mjs`},keywords:[`BRICKS`,`buttress`,`server`],license:`MIT`,dependencies:{"@elysiajs/cors":`^1.1.1`,"@elysiajs/node":`^1.4.2`,"@fugood/llama.node":`^1.7.0-rc.11`,"@fugood/whisper.node":`^1.0.18`,"@huggingface/gguf":`^0.3.2`,"@iarna/toml":`^3.0.0`,bytes:`^3.1.0`,elysia:`^1.4.19`,ms:`^2.1.1`,"node-machine-id":`^1.1.12`,zod:`^3.25.76`},devDependencies:{tsdown:`^0.20.1`,typescript:`^5.9.3`},gitHead:`d8a3bbd695e9118cf725b7226866d804bbe56bbe`};const Zr=()=>({version:Yr,name:Jr,description:Xr.description}),Qr=typeof process<`u`&&process.versions&&process.versions.node,$r=e=>new n({adapter:Qr?t():void 0,...e}),ei=a.Object({id:a.String(),name:a.String(),version:a.String(),generators:a.Array(a.Object({type:a.String()})),authentication:a.Object({required:a.Boolean(),type:a.Literal(`device-group`)})}),ti=({store:{serverInfo:e}})=>({id:e.id,name:e.name,version:e.version,generators:e.generators,authentication:e.authentication});var ni=e=>{let t=$r(),n=e.autodiscover.http?.path??`/buttress/info`;return t.get(n,ti,{response:ei}),t};const ri=typeof process<`u`&&process.versions!=null&&process.versions.node!=null;var ii=$r().post(`/buttress/upload`,async({body:{file:e},store:{config:t}})=>{let n=`${Date.now()}-${e.name.replace(/[^\dA-Za-z]/g,`_`)}`,r=_.join(t.server.temp_file_dir,n);try{return ri?await g(r,await e.stream()):await g(r,await e.arrayBuffer()),{ok:!0,filename:n}}catch(e){return{ok:!1,error:String(e)}}},{body:a.Object({file:a.File()}),response:a.Object({ok:a.Boolean(),filename:a.Optional(a.String()),error:a.Optional(a.String())})}).get(`/buttress/download/:filename`,async({params:{filename:e},store:{config:t},status:n})=>{let i=_.join(t.server.temp_file_dir,e);return _.relative(t.server.temp_file_dir,i).includes(`..`)?(n(400),`Invalid file path`):r(i)},{params:a.Object({filename:a.String()})});const ai=_.dirname(D(import.meta.url)),oi=async()=>{let e=[_.join(ai,`..`,`public`,`status.html`),_.join(ai,`..`,`..`,`public`,`status.html`)];return(await Promise.all(e.map(e=>c.access(e).then(()=>e,()=>null)))).find(e=>e!==null)??null},si=e=>{let{status:t}=e;return t?.getFullStatus?t.getFullStatus():{timestamp:new Date().toISOString(),ggmlLlm:{generators:[],history:{}},ggmlStt:{generators:[],history:{}}}},ci=async()=>{let e=await oi();if(!e)return console.error(`[Status] Failed to find status.html in candidate paths`),new Response(`Status page not found`,{status:404,headers:{"Content-Type":`text/plain`}});try{let t=await c.readFile(e,`utf-8`);return new Response(t,{headers:{"Content-Type":`text/html; charset=utf-8`}})}catch(e){return console.error(`[Status] Failed to serve status page:`,e),new Response(`Status page not found`,{status:404,headers:{"Content-Type":`text/plain`}})}};var li=$r().get(`/status`,ci).get(`/status/`,ci).get(`/buttress/status`,({store:{backend:e}})=>si(e));const ui=[`ggml-llm`,`mlx-llm`],di=new Map;function fi(e,t){return t===`mlx-llm`?e.mlxLlm:e.ggmlLlm}async function pi(e,t,n,r=`[LLM]`){let i=(t.generators||[]).filter(e=>ui.includes(e.type));if(i.length===0)throw Error(`No LLM generator configured. Add a [[generators]] with type = "ggml-llm" or "mlx-llm" to your config.`);let a=i[0],o=n||a.model?.repo_id;if(n){let e=i.find(e=>e.model?.repo_id===n);e&&(a=e)}else o=a.model?.repo_id;let s=a.type||`ggml-llm`,c=o,l=di.get(c);if(l?.initialized)return l;let{generators:u,server:d,...f}=t.global||{},p={...f,...a,model:{...a.model,repo_id:o}};console.log(`${r} Creating ${s} generator for ${c}`);let{id:m}=await e.startGenerator(s,p),h={id:m,type:s,config:p,repoId:o,initialized:!1};return di.set(c,h),await fi(e,s).initContext(m,{}),h.initialized=!0,console.log(`${r} Generator ready: ${c}`),h}function mi(e){let t=e.timings||{},n=t.prompt_n??t.promptN??0,r=t.cache_n??t.cacheN??0,i=t.predicted_n??t.predictedN??0;return{promptTokens:n||e.prompt_tokens||e.promptTokens||0,cachedTokens:r,completionTokens:i||e.tokens_evaluated||e.tokensEvaluated||e.tokens_predicted||e.tokensPredicted||0}}function hi(e){let{promptTokens:t,cachedTokens:n,completionTokens:r}=mi(e),i=t+n;return{prompt_tokens:i,completion_tokens:r,total_tokens:i+r}}const gi=()=>`chatcmpl-${Date.now()}-${Math.random().toString(36).slice(2,9)}`;async function _i(e,t,n,r){let i=e.getReader(),a=``,o=null,s=null,c=`stop`,l={prompt_tokens:0,completion_tokens:0,total_tokens:0};try{let e=!1;for(;!e;){let n=await i.read();if({done:e}=n,e)break;let{event:r,data:u}=n.value;if(r===`token`)u.content!=null&&(a=u.content),u.reasoning_content!=null&&(o=u.reasoning_content);else if(r===`result`)u.content==null?u.text&&(a=u.text):a=u.content,u.reasoning_content!=null&&(o=u.reasoning_content),u.tool_calls?.length>0?(s=u.tool_calls.map((e,n)=>({id:e.id||`call_${t}_${n}`,type:`function`,function:{name:e.function?.name||``,arguments:e.function?.arguments||``}})),c=`tool_calls`):c=u.interrupted?`length`:`stop`,l=hi(u);else if(r===`error`)throw Error(u.message)}}finally{i.cancel().catch(()=>{})}let u={role:`assistant`,content:a||null};return o&&(u.reasoning_content=o),s&&(u.tool_calls=s),{id:t,object:`chat.completion`,created:n,model:r,choices:[{index:0,message:u,finish_reason:c}],usage:l}}function vi({global:e}){let t=$r({prefix:`/oai-compat`});return t.use(M({origin:e?.openai_compat?.cors_allowed_origins??!1,methods:[`GET`,`POST`,`OPTIONS`],allowedHeaders:[`Content-Type`,`Authorization`],maxAge:86400,preflight:!0})),t.get(`/v1/models`,({store:e})=>{let{config:t}=e,n=(t.generators||[]).filter(e=>ui.includes(e.type)).map(e=>({id:e.model?.repo_id||e.type,object:`model`,created:Math.floor(Date.now()/1e3),owned_by:`local`}));return n.length===0&&n.push({id:`ggml-llm`,object:`model`,created:Math.floor(Date.now()/1e3),owned_by:`local`}),{object:`list`,data:n}}),t.post(`/v1/chat/completions`,async function*({body:e,set:t,store:n}){let{config:r,backend:a}=n,{messages:o=[],stream:s=!1,model:c,tools:l,temperature:u,stop:d,top_p:f,max_tokens:p,presence_penalty:m,frequency_penalty:h,tool_choice:g,stream_options:_,enable_thinking:v}=e;if(!o||o.length===0)return t.status=400,{error:{message:`messages is required and must not be empty`,type:`invalid_request_error`}};try{let e=await pi(a,r,c,`[OpenAI]`),t=gi(),n=Math.floor(Date.now()/1e3),y=e.repoId||`ggml-llm`,b={reasoning_format:`auto`,messages:o,jinja:!0,add_generation_prompt:!0};u!=null&&(b.temperature=u),f!=null&&(b.top_p=f),p!=null&&(b.n_predict=p),d!=null&&(b.stop=Array.isArray(d)?d:[d]),m!=null&&(b.presence_penalty=m),h!=null&&(b.frequency_penalty=h),l!=null&&(b.tools=l),g!=null&&(b.tool_choice=g),b.enable_thinking=v??!1;let x=await fi(a,e.type).completion(e.id,{options:b});if(!s)return await _i(x,t,n,y);let S=_?.include_usage===!0,C=x.getReader(),w=``,T=``,E=new Map,D=new Map;try{let e=!1;for(;!e;){let r=await C.read();if({done:e}=r,e)break;let{event:a,data:o}=r.value;if(a===`token`){let e={};if(o.content!=null){let t=o.content;t.length>w.length&&(e.content=t.slice(w.length),w=t)}if(o.reasoning_content!=null){let t=o.reasoning_content;t.length>T.length&&(e.reasoning_content=t.slice(T.length),T=t)}if(o.tool_calls?.length>0){let n=[];o.tool_calls.forEach((e,r)=>{let i={index:r};D.has(r)||(D.set(r,e.id||`call_${t}_${r}`),i.id=D.get(r),i.type=`function`);let a=e.function?.arguments||``,o=E.get(r)||``,s={};!E.has(r)&&e.function?.name&&(s.name=e.function.name),a.length>o.length&&(s.arguments=a.slice(o.length),E.set(r,a)),Object.keys(s).length>0?(i.function=s,n.push(i)):i.id&&(i.function={name:e.function?.name||``,arguments:``},n.push(i))}),n.length>0&&(e.tool_calls=n)}Object.keys(e).length>0&&(yield i({data:JSON.stringify({id:t,object:`chat.completion.chunk`,created:n,model:y,choices:[{index:0,delta:e,finish_reason:null}]})}))}else if(a===`result`){let e=`stop`;o.tool_calls?.length>0||D.size>0?e=`tool_calls`:o.interrupted&&(e=`length`);let r={id:t,object:`chat.completion.chunk`,created:n,model:y,choices:[{index:0,delta:{},finish_reason:e}]};S&&(r.usage=hi(o)),yield i({data:JSON.stringify(r)})}else a===`error`&&(yield i({data:JSON.stringify({error:{message:o.message,type:`server_error`}})}))}yield i({data:`[DONE]`})}finally{C.cancel().catch(()=>{})}}catch(e){return console.error(`[OpenAI] Chat completion error:`,e),t.status=500,{error:{message:e.message||`Internal server error`,type:`server_error`}}}},{body:a.Object({model:a.Optional(a.String()),messages:a.Array(a.Any()),stream:a.Optional(a.Boolean()),temperature:a.Optional(a.Number()),top_p:a.Optional(a.Number()),max_tokens:a.Optional(a.Number()),stop:a.Optional(a.Union([a.String(),a.Array(a.String())])),presence_penalty:a.Optional(a.Number()),frequency_penalty:a.Optional(a.Number()),tools:a.Optional(a.Array(a.Any())),tool_choice:a.Optional(a.Any()),stream_options:a.Optional(a.Object({include_usage:a.Optional(a.Boolean())})),enable_thinking:a.Optional(a.Boolean())})}),t}const yi=()=>`msg_${Date.now()}${Math.random().toString(36).slice(2,11)}`;function bi(e){let t={},n=[];if(e.system!=null){let t=``;if(typeof e.system==`string`)t=e.system;else if(Array.isArray(e.system))for(let n of e.system)n?.type===`text`&&typeof n.text==`string`&&(t+=n.text);t&&n.push({role:`system`,content:t})}if(!Array.isArray(e.messages))throw Error(`'messages' is required and must be an array`);for(let t of e.messages){let e=t?.role||`user`;if(t?.content==null){if(e===`assistant`)continue;n.push(t);continue}if(typeof t.content==`string`){n.push({role:e,content:t.content});continue}if(!Array.isArray(t.content)){n.push(t);continue}let r=[],i=[],a=[],o=``,s=!1;for(let e of t.content){let t=e?.type||``;if(t===`text`)i.push({type:`text`,text:e.text||``});else if(t===`thinking`)o+=e.thinking||``;else if(t===`image`){let t=e.source||{};if(t.type===`base64`){let e=t.media_type||`image/jpeg`,n=t.data||``;i.push({type:`image_url`,image_url:{url:`data:${e};base64,${n}`}})}else t.type===`url`&&i.push({type:`image_url`,image_url:{url:t.url||``}})}else if(t===`tool_use`)r.push({id:e.id||``,type:`function`,function:{name:e.name||``,arguments:JSON.stringify(e.input??{})}}),s=!0;else if(t===`tool_result`){let t=e.tool_use_id||``,n=``,r=e.content;if(typeof r==`string`)n=r;else if(Array.isArray(r))for(let e of r)e?.type===`text`&&(n+=e.text||``);a.push({role:`tool`,tool_call_id:t,content:n})}}if(i.length>0||s||o){let t={role:e};i.length>0?t.content=i:(s||o)&&(t.content=``),r.length>0&&(t.tool_calls=r),o&&(t.reasoning_content=o),n.push(t)}for(let e of a)n.push(e)}if(t.messages=n,Array.isArray(e.tools)&&(t.tools=e.tools.map(e=>({type:`function`,function:{name:e.name||``,description:e.description||``,parameters:e.input_schema||{}}}))),e.tool_choice&&typeof e.tool_choice==`object`){let n=e.tool_choice.type;n===`auto`?t.tool_choice=`auto`:n===`any`||n===`tool`?t.tool_choice=`required`:n===`none`&&(t.tool_choice=`none`)}else Array.isArray(t.tools)&&t.tools.length>0&&(t.tool_choice=`auto`);e.stop_sequences!=null&&(t.stop=Array.isArray(e.stop_sequences)?e.stop_sequences:[e.stop_sequences]),t.max_tokens=e.max_tokens??4096;for(let n of[`temperature`,`top_p`,`top_k`,`stream`])e[n]!=null&&(t[n]=e[n]);return e.thinking&&typeof e.thinking==`object`&&e.thinking.type===`enabled`&&(t.enable_thinking=!0,e.thinking.budget_tokens!=null&&(t.thinking_budget_tokens=e.thinking.budget_tokens)),t}function xi(e,t){return t?`tool_use`:e.stopping_word||e.stoppingWord?`stop_sequence`:e.interrupted||e.truncated?`max_tokens`:`end_turn`}function Si(e){let{promptTokens:t,cachedTokens:n,completionTokens:r}=mi(e);return{cache_read_input_tokens:n,input_tokens:t,output_tokens:r}}async function Ci(e,t,n){let r=e.getReader(),i=``,a=``,o=[],s={cache_read_input_tokens:0,input_tokens:0,output_tokens:0},c=`end_turn`,l=null;try{let e=!1;for(;!e;){let t=await r.read();if({done:e}=t,e)break;let{event:n,data:u}=t.value;if(n===`token`)u.content!=null&&(i=u.content),u.reasoning_content!=null&&(a=u.reasoning_content);else if(n===`result`)u.content==null?u.text&&u.reasoning_content==null&&(i=u.text):i=u.content,u.reasoning_content!=null&&(a=u.reasoning_content),Array.isArray(u.tool_calls)&&(o=u.tool_calls),s=Si(u),l=u.stopping_word||u.stoppingWord||null,c=xi(u,o.length>0);else if(n===`error`)throw Error(u.message||`completion error`)}}finally{r.cancel().catch(()=>{})}let u=[];a&&u.push({type:`thinking`,thinking:a,signature:``}),i&&u.push({type:`text`,text:i});for(let e of o){let t={};try{t=JSON.parse(e.function?.arguments||`{}`)}catch{t={}}u.push({type:`tool_use`,id:e.id||`toolu_${Math.random().toString(36).slice(2,11)}`,name:e.function?.name||``,input:t})}return{id:t,type:`message`,role:`assistant`,content:u,model:n,stop_reason:c,stop_sequence:l,usage:s}}function wi({global:e}){let t=$r({prefix:`/anthropic-messages`});return t.use(M({origin:e?.anthropic_messages?.cors_allowed_origins??!1,methods:[`GET`,`POST`,`OPTIONS`],allowedHeaders:[`Content-Type`,`Authorization`,`x-api-key`,`anthropic-version`],maxAge:86400,preflight:!0})),t.post(`/v1/messages`,async function*({body:e,set:t,store:n}){let{config:r,backend:a}=n,o=e;try{if(!Array.isArray(o.messages)||o.messages.length===0)return t.status=400,{type:`error`,error:{type:`invalid_request_error`,message:`messages is required and must not be empty`}};let e=bi(o),n=await pi(a,r,o.model,`[Anthropic]`),s=yi(),c=n.repoId||`ggml-llm`,l={reasoning_format:`auto`,messages:e.messages,jinja:!0,add_generation_prompt:!0};e.temperature!=null&&(l.temperature=e.temperature),e.top_p!=null&&(l.top_p=e.top_p),e.top_k!=null&&(l.top_k=e.top_k),e.max_tokens!=null&&(l.n_predict=e.max_tokens),e.stop!=null&&(l.stop=e.stop),e.tools!=null&&(l.tools=e.tools),e.tool_choice!=null&&(l.tool_choice=e.tool_choice),l.enable_thinking=e.enable_thinking??!1,e.thinking_budget_tokens!=null&&(l.thinking_budget_tokens=e.thinking_budget_tokens);let u=await fi(a,n.type).completion(n.id,{options:l});if(!o.stream)return await Ci(u,s,c);let d=u.getReader(),f=``,p=``,m=new Map,h=new Map,g=new Map,_=new Set,v=!1,y=!1,b=0,x=0,S={cache_read_input_tokens:0,input_tokens:0,output_tokens:0},C=`end_turn`,w=null,T=!1,E=e=>(v?1:0)+(y?1:0)+e;try{let e=!1;for(;!e;){let t=await d.read();if({done:e}=t,e)break;let{event:n,data:r}=t.value;if(n===`token`){if(!T){let e=Si(r);yield i({event:`message_start`,data:JSON.stringify({type:`message_start`,message:{id:s,type:`message`,role:`assistant`,content:[],model:c,stop_reason:null,stop_sequence:null,usage:e}})}),T=!0}if(r.reasoning_content!=null){let e=r.reasoning_content;e.length>p.length&&(v||(yield i({event:`content_block_start`,data:JSON.stringify({type:`content_block_start`,index:0,content_block:{type:`thinking`,thinking:``}})}),v=!0,b=1),yield i({event:`content_block_delta`,data:JSON.stringify({type:`content_block_delta`,index:0,delta:{type:`thinking_delta`,thinking:e.slice(p.length)}})}),p=e)}if(r.content!=null){let e=r.content;e.length>f.length&&(y||=(yield i({event:`content_block_start`,data:JSON.stringify({type:`content_block_start`,index:b,content_block:{type:`text`,text:``}})}),!0),yield i({event:`content_block_delta`,data:JSON.stringify({type:`content_block_delta`,index:b,delta:{type:`text_delta`,text:e.slice(f.length)}})}),f=e)}if(Array.isArray(r.tool_calls)&&r.tool_calls.length>0){for(let e=0;e<r.tool_calls.length;e+=1){let t=r.tool_calls[e],n=E(e),a=t?.function?.arguments||``,o=m.get(e)||``;if(!_.has(e)){let r=t?.id||`toolu_${s}_${e}`,a=t?.function?.name||g.get(e)||``;h.set(e,r),g.set(e,a),_.add(e),yield i({event:`content_block_start`,data:JSON.stringify({type:`content_block_start`,index:n,content_block:{type:`tool_use`,id:r,name:a,input:{}}})})}a.length>o.length&&(yield i({event:`content_block_delta`,data:JSON.stringify({type:`content_block_delta`,index:n,delta:{type:`input_json_delta`,partial_json:a.slice(o.length)}})}),m.set(e,a))}x=r.tool_calls.length}}else if(n===`result`){if(!T){let e=Si(r);yield i({event:`message_start`,data:JSON.stringify({type:`message_start`,message:{id:s,type:`message`,role:`assistant`,content:[],model:c,stop_reason:null,stop_sequence:null,usage:e}})}),T=!0}Array.isArray(r.tool_calls)&&(x=Math.max(x,r.tool_calls.length)),S=Si(r),w=r.stopping_word||r.stoppingWord||null,C=xi(r,_.size>0)}else if(n===`error`){yield i({event:`error`,data:JSON.stringify({type:`error`,error:{type:`api_error`,message:r.message||`completion error`}})});return}}v&&(yield i({event:`content_block_delta`,data:JSON.stringify({type:`content_block_delta`,index:0,delta:{type:`signature_delta`,signature:``}})}),yield i({event:`content_block_stop`,data:JSON.stringify({type:`content_block_stop`,index:0})})),y&&(yield i({event:`content_block_stop`,data:JSON.stringify({type:`content_block_stop`,index:b})}));for(let e of[..._].sort((e,t)=>e-t))yield i({event:`content_block_stop`,data:JSON.stringify({type:`content_block_stop`,index:E(e)})});yield i({event:`message_delta`,data:JSON.stringify({type:`message_delta`,delta:{stop_reason:C,stop_sequence:w},usage:{output_tokens:S.output_tokens}})}),yield i({event:`message_stop`,data:JSON.stringify({type:`message_stop`})})}finally{d.cancel().catch(()=>{})}}catch(e){return console.error(`[Anthropic] Messages error:`,e),t.status=500,{type:`error`,error:{type:`api_error`,message:e?.message||`Internal server error`}}}},{body:a.Object({model:a.Optional(a.String()),messages:a.Array(a.Any()),system:a.Optional(a.Any()),max_tokens:a.Optional(a.Number()),stream:a.Optional(a.Boolean()),temperature:a.Optional(a.Number()),top_p:a.Optional(a.Number()),top_k:a.Optional(a.Number()),stop_sequences:a.Optional(a.Union([a.String(),a.Array(a.String())])),tools:a.Optional(a.Array(a.Any())),tool_choice:a.Optional(a.Any()),thinking:a.Optional(a.Any()),metadata:a.Optional(a.Any())})}),t.post(`/v1/messages/count_tokens`,async({body:e,set:t,store:n})=>{let{config:r,backend:i}=n,a=e;try{let e=bi(a),t=await pi(i,r,a.model,`[Anthropic]`),n=fi(i,t.type),o={messages:e.messages,add_generation_prompt:!0,jinja:!0};e.tools!=null&&(o.tools=e.tools);let s=await n.applyChatTemplate(t.id,o),c=typeof s==`string`?s:s?.prompt||``,l=await n.tokenize(t.id,{text:c,add_special:!0,parse_special:!0});return{input_tokens:(Array.isArray(l)?l:l?.tokens||[]).length}}catch(e){return console.error(`[Anthropic] count_tokens error:`,e),t.status=500,{type:`error`,error:{type:`api_error`,message:e?.message||`Internal server error`}}}},{body:a.Object({model:a.Optional(a.String()),messages:a.Array(a.Any()),system:a.Optional(a.Any()),tools:a.Optional(a.Array(a.Any())),tool_choice:a.Optional(a.Any())})}),t.get(`/v1/models`,({store:e})=>{let{config:t}=e,n=(t.generators||[]).filter(e=>ui.includes(e.type)).map(e=>{let t=e.model?.repo_id||e.type;return{id:t,type:`model`,display_name:t,created_at:new Date().toISOString()}});return n.length===0&&n.push({id:`ggml-llm`,type:`model`,display_name:`ggml-llm`,created_at:new Date().toISOString()}),{data:n,has_more:!1,first_id:n[0]?.id,last_id:n.at(-1)?.id}}),t}const Ti=(e={},t={})=>{let n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&typeof t==`object`&&!Array.isArray(t)?n[e]=Ti(n[e]||{},t):n[e]=t}),n},Ei=e=>e&&typeof e==`object`?structuredClone(e):null,Di=(e,t)=>Ti(Ei(e)||{},Ei(t)||{}),Oi=(e,t)=>Ti(structuredClone(e.global),t||{}),ki=(e,t,n,r)=>{if(e.generators.length>0){let i=e.generators.filter(e=>e?.type===n);if(i.length>0&&r){let a=i.find(e=>t.getModelIdentifier(n,e)===r);if(a)return Oi(e,a)}}return Object.keys(e.global).length>0?Oi(e,{}):null},Ai={udp:{port:8089,announcements:{enabled:!0,interval:5e3},requests:{enabled:!0,responseDelay:100}},http:{enabled:!0,path:`/buttress/info`,cors:!0}},ji=e=>e?e===!0?{...Ai}:Ti(Ai,e):null,Mi=(e,t)=>{if(!e.generators||e.generators.length===0)return t.map(e=>({type:e}));let n=new Set;return e.generators.forEach(e=>{e.type&&n.add(e.type)}),n.size===0?t.map(e=>({type:e})):Array.from(n).map(e=>({type:e}))},Ni=(e,t,n)=>e===void 0?n:typeof e==`number`?e:t(e)??n,Pi=6e4,Fi=1024*1024*50,Ii=e=>{let t=N.machineIdSync(),n=Ti({server:{id:`buttress-${t}`,name:`Buttress Server (${t.slice(-8)})`,port:2080,temp_file_dir:_.join(v.tmpdir(),`.buttress`),session_timeout:Pi,max_body_size:Fi},autodiscover:!1},Ei(e)||{}),r=Array.isArray(n.generators)?n.generators:[],{server:i,generators:a,autodiscover:o,...s}=n;return{autodiscover:ji(o),server:{id:i.id,name:i.name,port:i.port,log_level:i.log_level,temp_file_dir:i.temp_file_dir,max_body_size:Ni(i.max_body_size,w.parse,Fi),session_timeout:Ni(i.session_timeout,P,Pi)},global:s,generators:r}},Li={getCapabilities:j.tuple([j.object({type:j.string().optional().default(`ggml-llm`),config:j.any().optional(),currentClientCapabilities:j.any().optional(),options:j.any().optional()}).nullable().optional()]),startGenerator:j.tuple([j.string(),j.any().optional()]),finalizeGenerator:j.tuple([j.string()])};var Ri={async getCapabilities({backend:e,config:t},n=null){console.log(`[Server] Get Capabilities:`,n);let{type:r=`ggml-llm`,config:i,currentClientCapabilities:a=null,options:o={}}=n||{type:`ggml-llm`},s=Ei(i),c=Di(ki(t,e,r,e.getModelIdentifier(r,s)),i);if(Object.keys(c).length===0)throw Error(`Buttress server missing generator configuration`);return c.backend=c.backend||{},c.backend.type||(c.backend.type=r),e.getCapabilities(r,a,{...o,config:c})},async startGenerator({backend:e,config:t,session:n},r,i){console.log(`[Server] Start Generator:`,r,i);let a=Ei(i),o=Di(ki(t,e,r,e.getModelIdentifier(r,a)),i);if(Object.keys(o).length===0)throw Error(`Buttress server missing generator configuration`);o.backend=o.backend||{},o.backend.type||(o.backend.type=r);let s=await e.startGenerator(r,o);return n.generators.add(s.id),s},async finalizeGenerator({backend:e,session:t},n){return console.log(`[Server] Finalize Generator:`,n),t.generators.delete(n),e.finalizeGenerator(n)}};const zi={initContext:j.tuple([j.string(),j.any().optional()]),completion:j.tuple([j.string(),j.any().optional()]),tokenize:j.tuple([j.string(),j.any()]),detokenize:j.tuple([j.string(),j.any()]),applyChatTemplate:j.tuple([j.string(),j.any()]),releaseContext:j.tuple([j.string()])};function Bi(e){return function({backend:t,session:n},r,i){return new s({async start(a){try{let o=await e(t).initContext(r,{...i,onProgress:e=>{a.enqueue({event:`progress`,data:{progress:e}})}});n.initializedContexts.add(r),await new Promise(e=>setTimeout(e,1e3));let{download:s,...c}=o||{};a.enqueue({event:`result`,data:{result:c}}),a.close()}catch(e){a.error(e)}}})}}function Vi(e,t){return async function({backend:n,session:r},i,a){return console.log(`[Server] ${t}:`,{id:i,force:a}),r.initializedContexts.has(i)?(r.initializedContexts.delete(i),e(n).releaseContext(i,{force:a})):(console.log(`[Server] ${t} skipped - not initialized by this session:`,{id:i}),{released:!1,skipped:!0})}}function Hi(e,t){return{initContext:Bi(e),completion({backend:n},r,i){return console.log(`[Server] ${t}Completion:`,{id:r,property:i}),e(n).completion(r,i)},async tokenize({backend:n},r,i){return console.log(`[Server] ${t}Tokenize:`,{id:r,property:i}),e(n).tokenize(r,i)},async detokenize({backend:n},r,i){return console.log(`[Server] ${t}Detokenize:`,{id:r,property:i}),e(n).detokenize(r,i)},async applyChatTemplate({backend:n},r,i){return console.log(`[Server] ${t}Apply Chat Template:`,{id:r,property:i}),e(n).applyChatTemplate(r,i)},releaseContext:Vi(e,`${t}Release Context`)}}var Ui=Hi(e=>e.ggmlLlm,``);const Wi={initContext:j.tuple([j.string(),j.any().optional()]),transcribe:j.tuple([j.string(),j.string(),j.any().optional()]),transcribeData:j.tuple([j.string(),j.union([j.instanceof(Buffer),j.instanceof(Uint8Array)]),j.any().optional()]),releaseContext:j.tuple([j.string()])},Gi=e=>e.ggmlStt,Ki={common:Ri,ggmlLlm:Ui,ggmlStt:{initContext:Bi(Gi),async transcribe({backend:e,config:{server:t}},n,r,i){return console.log(`[Server] Transcribe:`,{id:n,audioPath:r,options:i}),e.ggmlStt.transcribe(n,{audioPath:_.join(t.temp_file_dir,r),options:i})},async transcribeData({backend:e},t,n,r){return console.log(`[Server] Transcribe Data:`,{id:t,audioDataLength:n?.length||0,options:r}),e.ggmlStt.transcribeData(t,{audioData:n,options:r})},releaseContext:Vi(Gi,`Release STT Context`)},mlxLlm:Hi(e=>e.mlxLlm,`MLX `)},qi={common:Li,ggmlLlm:zi,ggmlStt:Wi,mlxLlm:zi};var Ji=Ki;const Yi=e=>{try{return JSON.parse(e,(e,t)=>{if(!t)return t;if(t?.type===`Buffer`&&t?.data)return F.from(t.data,`base64`);if(t?.type===`Uint8Array`&&t?.data){let e=F.from(t.data,`base64`);return e.buffer.slice(e.byteOffset,e.byteOffset+e.byteLength)}return t?.type===`Error`&&t?.name&&t?.message?Error(t.name,t.message):t})}catch{return e}},Xi=e=>{try{return JSON.stringify(e,(e,t)=>t instanceof Error?{type:`Error`,name:t.name,message:t.message}:t instanceof F?{type:`Buffer`,data:t.toString(`base64`)}:t instanceof Uint8Array?{type:`Uint8Array`,data:F.from(t).toString(`base64`)}:t)}catch{return e}};var Zi=class{name=`udp`;socket=null;announcementTimer=null;config;getServerInfo;port;constructor(e,t){this.config=e,this.getServerInfo=t,this.port=e.port??8089}async start(){if(this.socket=re.createSocket({type:`udp4`,reuseAddr:!0}),this.socket.on(`message`,(e,t)=>{this.handleMessage(e,t)}),this.socket.on(`error`,e=>{console.error(`[Autodiscover UDP] Socket error:`,e.message)}),await new Promise((e,t)=>{this.socket.bind(this.port,()=>{this.socket.setBroadcast(!0),console.log(`[Autodiscover UDP] Listening on port ${this.port}`),e()}),this.socket.once(`error`,t)}),this.config.announcements.enabled){let e=this.config.announcements.interval??5e3;this.announcementTimer=setInterval(()=>{this.sendAnnouncement()},e),this.sendAnnouncement()}}async stop(){this.announcementTimer&&=(clearInterval(this.announcementTimer),null),this.socket&&=(await new Promise(e=>{this.socket.close(()=>e())}),null)}handleMessage(e,t){try{let n=JSON.parse(e.toString());if(n.t===`QUERY`&&this.config.requests.enabled){let e=n.d,r=this.config.requests.responseDelay??0,i=r>0?Math.random()*r:0;setTimeout(()=>{this.sendResponse(e.id,t)},i)}}catch{}}sendAnnouncement(){if(!this.socket)return;let e={t:`ANNOUNCE`,v:`1.0`,d:{info:this.getServerInfo()}},t=Buffer.from(JSON.stringify(e));this.socket.send(t,0,t.length,this.port,`255.255.255.255`,e=>{e&&console.error(`[Autodiscover UDP] Announcement error:`,e.message)})}sendResponse(e,t){if(!this.socket)return;let n={t:`RESPONSE`,v:`1.0`,d:{request_id:e,info:this.getServerInfo()}},r=Buffer.from(JSON.stringify(n));this.socket.send(r,0,r.length,t.port,t.address,e=>{e&&console.error(`[Autodiscover UDP] Response error:`,e.message)})}},Qi=class{transports=[];started=!1;constructor(e,t){this.config=e,this.getServerInfo=t,(e.udp?.announcements?.enabled||e.udp?.requests?.enabled)&&this.transports.push(new Zi(e.udp,t))}async start(){this.started||=((await Promise.allSettled(this.transports.map(e=>e.start()))).forEach((e,t)=>{e.status===`rejected`&&console.error(`[Autodiscover] Failed to start ${this.transports[t].name}:`,e.reason)}),!0)}async stop(){this.started&&=(await Promise.allSettled(this.transports.map(e=>e.stop())),!1)}};const $i=()=>{let e=v.networkInterfaces();return Object.values(e).flat().find(e=>e?.family===`IPv4`&&!e?.internal)?.address||null},$=Zr(),ea=e=>{if(!e)return{repoId:null,filename:null};let[t,n]=e.split(`:`);return{repoId:t,filename:n||null}};async function ta({modelIds:e=[],defaultConfig:t=null}={}){let n=[];console.log(`${$.name} v${$.version}`),console.log(`Generating model capabilities comparison...
 `),n.push(`${$.name} v${$.version}`),n.push(`## Model Capabilities Comparison
 `),(!e||e.length===0)&&(console.error(`Error: No model IDs provided`),process.exit(1));try{let r=(e={},t={})=>{let n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&typeof t==`object`&&!Array.isArray(t)?n[e]=r(n[e]||{},t):n[e]=t}),n},{server:i,generators:a=[],...o}=t||{},s=e=>r(structuredClone(o),e||{}),c=e=>{if(Array.isArray(a)&&a.length>0){let t=a.filter(e=>e?.type===`ggml-llm`);if(t.length>0&&e){let n=t.find(t=>t.model?.repo_id===e);if(n)return s(n)}}return Object.keys(o).length>0?s({}):null},l=[];for(let t=0;t<e.length;t+=1){let n=e[t];console.log(`[${t+1}/${e.length}] Analyzing ${n}...`);let r=c(n);r={...r||{},model:{...o.runtime,...r?.model||{},repo_id:n}};let i=await J(`ggml-llm`,null,{config:r,includeBreakdown:!0});l.push({modelId:n,capabilities:i,modelInfo:i.buttress?.selected||null,modelConfig:i.modelConfig||null})}let u=e=>e?(e/1024/1024/1024).toFixed(2):`N/A`,d=e=>e?`✅`:`🚫`;n.push(`| Model ID | Size (GB) | Context Size | KV Cache Size (GB) | Recurrent Mem (GB) | Total Required Memory (GB) | Fits GPU (Full) | Fits CPU (Full) |`),n.push(`|----------|-----------|--------------|--------------------|--------------------|----------------------------|-----------------|-----------------|`),l.forEach(({modelId:e,modelInfo:t,modelConfig:r})=>{let i=u(t?.modelBytes),a=r?.nCtx||t?.kvInfo?.nCtxTrain||`N/A`,o=Ue(t),s=Number(a),c=t?.kvCacheBytes||(o&&Number.isFinite(s)&&s>0?o(s):o&&o(t?.kvInfo?.nCtxTrain||0))||null,l=u(c),f=t?.recurrentMemoryBytes||0,p=f>0?u(f):`-`,m=u(t?.modelBytes&&(c!=null||f>0)?t.modelBytes+(c||0)+f:t?.fit?.totalRequiredBytes),h=d(t?.fit?.fitsInGpu),g=d(t?.fit?.fitsInCpu);n.push(`| ${e} | ${i} | ${a} | ${l} | ${p} | ${m} | ${h} | ${g} |`);let _=t?.memoryLimitedCtx!=null||t?.limitedFit!=null,v=!t?.fit?.fitsInGpu||!t?.fit?.fitsInCpu;if(_&&v){let e=t?.memoryLimitedCtx||a,r=Number(e),s=t?.limitedKvCacheBytes||o&&Number.isFinite(r)&&r>0&&o(r)||null,c=u(s),h=u(t?.modelBytes&&(s!=null||f>0)?t.modelBytes+(s||0)+f:t?.limitedFit?.totalRequiredBytes),g=d(t?.limitedFit?.fitsInGpu),_=d(t?.limitedFit?.fitsInCpu);(e!==a||c!==l||h!==m)&&n.push(`| ↳ Limited | ${i} | ${e} | ${c} | ${p} | ${h} | ${g} | ${_} |`)}}),n.push(`
 ---`),n.push(`
@@ -99,9 +99,7 @@ Testing Options:
         file (e.g., ggml-llm-model-capabilities-YYYY-MM-DD.md)
 
 Environment Variables:
-  NODE_ENV                            Set to 'development' for dev mode
-  ENABLE_OPENAI_COMPAT_ENDPOINT       Set to '1' to enable OpenAI-compatible API
-  ENABLE_ANTHROPIC_MESSAGES_ENDPOINT  Set to '1' to enable Anthropic Messages API
+  NODE_ENV                      Set to 'development' for dev mode
 
 Examples:
   bricks-buttress
@@ -111,4 +109,4 @@ Examples:
   bricks-buttress --test-caps ggml-stt --test-caps-model-id BricksDisplay/whisper-ggml:ggml-small.bin
   bricks-buttress --test-caps mlx-llm --test-models-default
 `),process.exit(0));let e=process.argv.findIndex(e=>e===`--port`||e===`-p`),t=e>=0?Number(process.argv[e+1]):void 0,n=process.argv.findIndex(e=>e===`--config`||e===`-c`),r=n>=0?process.argv[n+1]:null,i=null;if(r){let e;if(r.includes(`
-`))e=r;else{let t=_.resolve(r);try{e=k.readFileSync(t,`utf8`)}catch(e){console.error(`Failed to read Buttress config at ${t}:`,e),process.exit(1)}}try{let t=A.parse(e);t.env&&typeof t.env==`object`&&(Object.entries(t.env).forEach(([e,t])=>{process.env[e]===void 0&&(process.env[e]=String(t))}),delete t.env),i=t}catch(e){console.error(`Failed to parse TOML config:`,e),process.exit(1)}}async function a(e){if(!e?.generators||!Array.isArray(e.generators))return;let t=e.generators.filter(e=>{if(!e.model?.download)return!1;let{type:t}=e;return!t||t!==`ggml-llm`&&t!==`ggml-stt`&&t!==`mlx-llm`?(console.warn(`[Download] Skipping unknown generator type: ${t}`),!1):!0});if(t.length===0)return;let{server:n,generators:r,...i}=e,a=t.map(e=>{let{type:t}=e,n=e.model?.repo_id;return console.log(`[Download] Starting pre-download for ${t}: ${n}`),qr(t,{...i,backend:e.backend||{},model:e.model||{},runtime:{...i.runtime,...e.runtime||{}}},{onProgress:()=>{},onComplete:({repoId:e,alreadyExists:t})=>{t?console.log(`[Download] Pre-download complete (already exists): ${e}`):console.log(`[Download] Pre-download complete: ${e}`)},onError:e=>{console.error(`[Download] Pre-download failed for ${n}:`,e.message)}})}),o=await Promise.all(a),s=o.filter(e=>e.started).length,c=o.filter(e=>e.alreadyExists).length,l=o.filter(e=>e.alreadyDownloading).length;console.log(`[Download] Pre-download summary: ${s} started, ${c} already exist, ${l} already downloading`)}let o=[`ggml-org/gpt-oss-20b-GGUF`,`ggml-org/gpt-oss-120b-GGUF`,`unsloth/Nemotron-3-Nano-30B-A3B-GGUF`,`unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF`,`unsloth/Qwen3.5-27B-GGUF`,`unsloth/gemma-4-26B-A4B-it-GGUF`,`unsloth/gemma-4-31B-it-GGUF`,`unsloth/GLM-4.7-Flash-GGUF`,`DevQuasar/MiniMaxAI.MiniMax-M2.5-GGUF`,`bartowski/Mistral-Nemo-Instruct-2407-GGUF`,`mistralai/Magistral-Small-2509-GGUF`,`mistralai/Ministral-3-14B-Reasoning-2512-GGUF`,`bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF`,`bartowski/mistralai_Devstral-2-123B-Instruct-2512-GGUF`,`ggml-org/gemma-3-12b-it-qat-GGUF`,`ggml-org/gemma-3-27b-it-qat-GGUF`,`unsloth/phi-4-GGUF`],s=[`BricksDisplay/whisper-ggml:ggml-small.bin`,`BricksDisplay/whisper-ggml:ggml-small-q8_0.bin`,`BricksDisplay/whisper-ggml:ggml-medium.bin`,`BricksDisplay/whisper-ggml:ggml-medium-q8_0.bin`,`BricksDisplay/whisper-ggml:ggml-large-v3-turbo.bin`,`BricksDisplay/whisper-ggml:ggml-large-v3-turbo-q8_0.bin`,`BricksDisplay/whisper-ggml:ggml-large-v3.bin`],c=[`mlx-community/Qwen3.5-27B-8bit`,`mlx-community/Qwen3.5-27B-4bit`,`mlx-community/Qwen3.5-35B-A3B-8bit`,`mlx-community/Qwen3.5-35B-A3B-4bit`,`mlx-community/Qwen3-235B-A22B-8bit`,`mlx-community/Qwen3-235B-A22B-4bit`,`mlx-community/GLM-4.7-Flash-8bit`,`mlx-community/GLM-4.7-Flash-4bit`,`mlx-community/MiniMax-M2.5-4bit`,`mlx-community/gpt-oss-120b-4bit`,`mlx-community/gemma-4-26b-a4b-it-8bit`,`mlx-community/gemma-4-26b-a4b-it-4bit`,`mlx-community/gemma-4-31b-it-8bit`,`mlx-community/gemma-4-31b-it-4bit`],l=process.argv.findIndex(e=>e===`--test-caps`);if(l>=0){let e=process.argv[l+1]||`ggml-llm`;e!==`ggml-llm`&&e!==`ggml-stt`&&e!==`mlx-llm`&&(console.error(`Only ggml-llm, ggml-stt, and mlx-llm backends are supported for testing capabilities`),process.exit(1));let t=process.argv.findIndex(e=>e===`--test-models`),n=process.argv.includes(`--test-models-default`);if(e===`mlx-llm`)if(t>=0){let e=process.argv[t+1];e||(console.error(`Error: --test-models requires a comma-separated list of model IDs`),process.exit(1)),await ia({modelIds:e.split(`,`).map(e=>e.trim()),defaultConfig:i})}else if(n)await ia({modelIds:c,defaultConfig:i});else{let e=process.argv.findIndex(e=>e===`--test-caps-model-id`);await aa({modelId:e>=0?process.argv[e+1]:null,defaultConfig:i})}else if(e===`ggml-stt`)if(t>=0){let e=process.argv[t+1];e||(console.error(`Error: --test-models requires a comma-separated list of model IDs`),process.exit(1)),await ra({modelIds:e.split(`,`).map(e=>e.trim()),defaultConfig:i})}else if(n)await ra({modelIds:s,defaultConfig:i});else{let e=process.argv.findIndex(e=>e===`--test-caps-model-id`);await oa({modelId:e>=0?process.argv[e+1]:null,defaultConfig:i})}else if(t>=0){let e=process.argv[t+1];e||(console.error(`Error: --test-models requires a comma-separated list of model IDs`),process.exit(1)),await ta({modelIds:e.split(`,`).map(e=>e.trim()),defaultConfig:i})}else if(n)await ta({modelIds:o,defaultConfig:i});else{let e=process.argv.findIndex(e=>e===`--test-caps-model-id`);await na({modelId:e>=0?process.argv[e+1]:null,defaultConfig:i})}}let u=Ii(i);t&&(u.server.port=t),u.server.port||(u.server.port=2080),_a({config:u,enableOpenAICompat:process.env.ENABLE_OPENAI_COMPAT_ENDPOINT===`1`,enableAnthropicMessages:process.env.ENABLE_ANTHROPIC_MESSAGES_ENDPOINT===`1`}).then(async({port:e,openaiEnabled:t,anthropicMessagesEnabled:n,autoDiscover:r})=>{let o=$i();console.log(`Buttress server listening on port ${e}`),console.log(`--------------------------------`),await ma(),console.log();let s={"ggml-llm":`LLM (GGML)`,"ggml-stt":`STT (GGML)`,"mlx-llm":`LLM (MLX)`};console.log(`Current supported Generators:`);let c=new Set((u?.generators||[]).map(e=>e.type).filter(Boolean));if(c.size===0)console.log(`- LLM (GGML)`),console.log(`- STT (GGML)`);else for(let e of c)console.log(`- ${s[e]||e}`);console.log(),console.log("Please configure `Buttress (Remote Inference)` in the Generator to connect to this server."),console.log(),console.log(`- Use http://${o}:${e} to connect to this server via LAN.`),console.log(`- Visit http://${o}:${e}/status to see status via LAN.`),console.log(),t?(console.log(`OpenAI-compatible API [EXPERIMENTAL]:`),console.log(`- Base URL: http://${o}:${e}/oai-compat/v1`),console.log(`- Chat completions: POST http://${o}:${e}/oai-compat/v1/chat/completions`),console.log(`- Models: GET http://${o}:${e}/oai-compat/v1/models`),console.log()):(console.log(`OpenAI-compatible API [EXPERIMENTAL]: disabled`),console.log(`  Set ENABLE_OPENAI_COMPAT_ENDPOINT=1 to enable`),console.log()),n?(console.log(`Anthropic Messages API [EXPERIMENTAL]:`),console.log(`- Base URL: http://${o}:${e}/anthropic-messages`),console.log(`- Messages: POST http://${o}:${e}/anthropic-messages/v1/messages`),console.log(`- Count tokens: POST http://${o}:${e}/anthropic-messages/v1/messages/count_tokens`),console.log()):(console.log(`Anthropic Messages API [EXPERIMENTAL]: disabled`),console.log(`  Set ENABLE_ANTHROPIC_MESSAGES_ENDPOINT=1 to enable`),console.log()),r&&(console.log(`Auto-discover enabled`),console.log()),i&&await a(i)}).catch(e=>{console.error(`Failed to start Buttress server:`,e),process.exitCode=1})};const{version:la,name:ua}=Zr(),da=async()=>{let e=`https://registry.npmjs.org/${ua}/latest`;try{let t=new AbortController,n=setTimeout(()=>t.abort(),3e3),r=await fetch(e,{headers:{Accept:`application/json`},signal:t.signal});return clearTimeout(n),r.ok&&(await r.json()).version||null}catch{return null}},fa=(e,t)=>{if(!t)return!1;let n=e.split(/[.-]/),r=t.split(/[.-]/);for(let e=0;e<Math.max(n.length,r.length);e+=1){let t=parseInt(n[e])||0,i=parseInt(r[e])||0;if(i>t)return!0;if(i<t)return!1}return!1},pa=e=>{console.log(``),console.log(`\x1B[33m╭─────────────────────────────────────────────────╮\x1B[0m`),console.log(`\x1B[33m│\x1B[0m  Update available! \x1B[2m%s\x1B[0m → \x1B[32m%s\x1B[0m`,la.padEnd(12),e.padEnd(12),`\x1B[33m│\x1B[0m`),console.log(`\x1B[33m│\x1B[0m                                                 \x1B[33m│\x1B[0m`),console.log(`\x1B[33m│\x1B[0m  Run to upgrade:                                \x1B[33m│\x1B[0m`),console.log(`\x1B[33m│\x1B[0m  \x1B[36mnpm install -g %s\x1B[0m      \x1B[33m│\x1B[0m`,ua.padEnd(27)),console.log(`\x1B[33m╰─────────────────────────────────────────────────╯\x1B[0m`),console.log(``)},ma=async()=>{try{let e=await da();e&&fa(la,e)&&pa(e)}catch{}},ha=typeof process<`u`&&process.versions&&process.versions.node,ga=async({backend:e,router:r,config:i,enableOpenAICompat:o,enableAnthropicMessages:l})=>{try{await c.mkdir(i.server.temp_file_dir,{recursive:!0})}catch{}let u=$i()||`0.0.0.0`,d={id:i.server.id,name:i.server.name,version:la,address:u,port:i.server.port,url:`http://${u}:${i.server.port}`,generators:Mi(i,i.generators.map(e=>e.type)),authentication:{required:!0,type:`device-group`}},f=new n({serve:{maxRequestBodySize:i.server.max_body_size},websocket:{idleTimeout:Math.ceil(i.server.session_timeout/1e3)},adapter:ha?t():void 0}).state({sessions:new Map,backend:e||Lr,config:i,serverInfo:d});r&&f.use(r),i.autodiscover?.http?.enabled&&f.use(ni(i)),f.use(ii),f.use(li),o&&f.use(vi(i)),l&&f.use(wi(i));let p={INVALID_REQUEST:-32600,INVALID_PARAMS:-32602,METHOD_NOT_FOUND:-32601,INTERNAL_ERROR:-32603},m=e=>e.id??e.raw?.id??e.remoteAddress;return f.ws(`/buttress/rpc`,{parse:(e,t)=>{if(typeof t==`string`)try{return JSON.parse(t)}catch{return e.send(JSON.stringify({jsonrpc:`2.0`,error:{code:p.INVALID_REQUEST,message:`Invalid request`},id:null})),null}return t},body:a.Object({jsonrpc:a.String(),method:a.String(),params:a.String(),id:a.String()}),open(e){let t=m(e);if(console.log(`[Request] New connection: ${t}`),!e.data.store.sessions.has(t))e.data.store.sessions.set(t,{streamReaders:new Map,generators:new Set,initializedContexts:new Set,timeout:null});else{let n=e.data.store.sessions.get(t);n&&(clearTimeout(n.timeout),n.timeout=null)}},async message(e,{id:t,method:n,params:r}){let i=m(e);console.log(`[Request] Received request from ${i}: ${n}`);let a=e.data.store.sessions.get(i),[o,c]=n.split(`.`),l=Ji[o]?.[c];if(!l){e.send(JSON.stringify({jsonrpc:`2.0`,error:{code:p.METHOD_NOT_FOUND,message:`Method not found`},id:t}));return}try{if(n===`cancel`){a.streamReaders.has(t)&&(a.streamReaders.get(t)?.cancel(),a.streamReaders.delete(t));return}if(n===`ping`){e.send(JSON.stringify({jsonrpc:`2.0`,result:`pong`,id:t}));return}let u=Yi(r),d=qi[o]?.[c],f=d?d.parse(u):u,p=await l({...e.data.store,peerId:i,session:a},...f);if(p instanceof s){let r=p.getReader();a.streamReaders.set(t,r),e.send(JSON.stringify({jsonrpc:`2.0`,result:{type:`stream`},id:t}));try{for(;;){let{value:n,done:i}=await r.read();if(i)break;let{event:a,data:o}=n;e.send(JSON.stringify({jsonrpc:`2.0`,method:`notification/${a}`,params:Xi(o),id:t}))}e.send(JSON.stringify({jsonrpc:`2.0`,method:`notification/_end`,id:t}))}catch(r){console.error(`[RPC] Stream error for ${n}:`,r),e.send(JSON.stringify({jsonrpc:`2.0`,method:`notification/_error`,params:Xi(r),id:t}))}a.streamReaders.delete(t)}else e.send(JSON.stringify({jsonrpc:`2.0`,result:Xi(p),id:t}))}catch(r){if(r instanceof ne){e.send(JSON.stringify({jsonrpc:`2.0`,error:{code:p.INVALID_PARAMS,message:`Invalid params`,data:r.issues},id:t}));return}console.error(`[RPC] Handler error for ${n}:`,r),e.send(JSON.stringify({jsonrpc:`2.0`,error:{code:p.INTERNAL_ERROR,message:String(r)},id:t}))}},async close(e){let t=m(e);console.log(`[Request] Connection closed: ${t}`);let{backend:n,sessions:r}=e.data.store,a=r.get(t);a&&(a.streamReaders.forEach(e=>e.cancel()),a.streamReaders.clear(),a.timeout=setTimeout(()=>{r.delete(t),console.log(`[Request] Session timed out: ${t}`),a.generators.forEach(e=>{n.finalizeGenerator(e)})},i.server.session_timeout))}}),{app:f,config:i}},_a=async({backend:e,router:t,config:n,enableOpenAICompat:r=!1,enableAnthropicMessages:i=!1})=>{let{app:a,config:o}=await ga({backend:e,router:t,config:n,enableOpenAICompat:r,enableAnthropicMessages:i}),{server:{port:s}}=o,c=[new Promise(e=>a.listen(s,e))],l=null;return o.autodiscover&&(l=new Qi(o.autodiscover,()=>a.store.serverInfo),c.push(l.start())),await Promise.all(c),{app:a,port:s,openaiEnabled:r,anthropicMessagesEnabled:i,autoDiscover:l}},va=[new URL(`index.mjs`,import.meta.url).pathname,new URL(`index.ts`,import.meta.url).pathname];(process.argv[1]?.endsWith(`/bricks-buttress`)||va.includes(process.argv[1]))&&await ca();export{ma as checkAndNotifyUpdates,da as checkForUpdates,fa as compareVersions,ga as createServer,pa as logUpdateMessage,Ii as processConfig,qr as startModelDownload,_a as startServer};
+`))e=r;else{let t=_.resolve(r);try{e=k.readFileSync(t,`utf8`)}catch(e){console.error(`Failed to read Buttress config at ${t}:`,e),process.exit(1)}}try{let t=A.parse(e);t.env&&typeof t.env==`object`&&(Object.entries(t.env).forEach(([e,t])=>{process.env[e]===void 0&&(process.env[e]=String(t))}),delete t.env),i=t}catch(e){console.error(`Failed to parse TOML config:`,e),process.exit(1)}}async function a(e){if(!e?.generators||!Array.isArray(e.generators))return;let t=e.generators.filter(e=>{if(!e.model?.download)return!1;let{type:t}=e;return!t||t!==`ggml-llm`&&t!==`ggml-stt`&&t!==`mlx-llm`?(console.warn(`[Download] Skipping unknown generator type: ${t}`),!1):!0});if(t.length===0)return;let{server:n,generators:r,...i}=e,a=t.map(e=>{let{type:t}=e,n=e.model?.repo_id;return console.log(`[Download] Starting pre-download for ${t}: ${n}`),qr(t,{...i,backend:e.backend||{},model:e.model||{},runtime:{...i.runtime,...e.runtime||{}}},{onProgress:()=>{},onComplete:({repoId:e,alreadyExists:t})=>{t?console.log(`[Download] Pre-download complete (already exists): ${e}`):console.log(`[Download] Pre-download complete: ${e}`)},onError:e=>{console.error(`[Download] Pre-download failed for ${n}:`,e.message)}})}),o=await Promise.all(a),s=o.filter(e=>e.started).length,c=o.filter(e=>e.alreadyExists).length,l=o.filter(e=>e.alreadyDownloading).length;console.log(`[Download] Pre-download summary: ${s} started, ${c} already exist, ${l} already downloading`)}let o=[`ggml-org/gpt-oss-20b-GGUF`,`ggml-org/gpt-oss-120b-GGUF`,`unsloth/Nemotron-3-Nano-30B-A3B-GGUF`,`unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF`,`unsloth/Qwen3.5-27B-GGUF`,`unsloth/gemma-4-26B-A4B-it-GGUF`,`unsloth/gemma-4-31B-it-GGUF`,`unsloth/GLM-4.7-Flash-GGUF`,`DevQuasar/MiniMaxAI.MiniMax-M2.5-GGUF`,`bartowski/Mistral-Nemo-Instruct-2407-GGUF`,`mistralai/Magistral-Small-2509-GGUF`,`mistralai/Ministral-3-14B-Reasoning-2512-GGUF`,`bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF`,`bartowski/mistralai_Devstral-2-123B-Instruct-2512-GGUF`,`ggml-org/gemma-3-12b-it-qat-GGUF`,`ggml-org/gemma-3-27b-it-qat-GGUF`,`unsloth/phi-4-GGUF`],s=[`BricksDisplay/whisper-ggml:ggml-small.bin`,`BricksDisplay/whisper-ggml:ggml-small-q8_0.bin`,`BricksDisplay/whisper-ggml:ggml-medium.bin`,`BricksDisplay/whisper-ggml:ggml-medium-q8_0.bin`,`BricksDisplay/whisper-ggml:ggml-large-v3-turbo.bin`,`BricksDisplay/whisper-ggml:ggml-large-v3-turbo-q8_0.bin`,`BricksDisplay/whisper-ggml:ggml-large-v3.bin`],c=[`mlx-community/Qwen3.5-27B-8bit`,`mlx-community/Qwen3.5-27B-4bit`,`mlx-community/Qwen3.5-35B-A3B-8bit`,`mlx-community/Qwen3.5-35B-A3B-4bit`,`mlx-community/Qwen3-235B-A22B-8bit`,`mlx-community/Qwen3-235B-A22B-4bit`,`mlx-community/GLM-4.7-Flash-8bit`,`mlx-community/GLM-4.7-Flash-4bit`,`mlx-community/MiniMax-M2.5-4bit`,`mlx-community/gpt-oss-120b-4bit`,`mlx-community/gemma-4-26b-a4b-it-8bit`,`mlx-community/gemma-4-26b-a4b-it-4bit`,`mlx-community/gemma-4-31b-it-8bit`,`mlx-community/gemma-4-31b-it-4bit`],l=process.argv.findIndex(e=>e===`--test-caps`);if(l>=0){let e=process.argv[l+1]||`ggml-llm`;e!==`ggml-llm`&&e!==`ggml-stt`&&e!==`mlx-llm`&&(console.error(`Only ggml-llm, ggml-stt, and mlx-llm backends are supported for testing capabilities`),process.exit(1));let t=process.argv.findIndex(e=>e===`--test-models`),n=process.argv.includes(`--test-models-default`);if(e===`mlx-llm`)if(t>=0){let e=process.argv[t+1];e||(console.error(`Error: --test-models requires a comma-separated list of model IDs`),process.exit(1)),await ia({modelIds:e.split(`,`).map(e=>e.trim()),defaultConfig:i})}else if(n)await ia({modelIds:c,defaultConfig:i});else{let e=process.argv.findIndex(e=>e===`--test-caps-model-id`);await aa({modelId:e>=0?process.argv[e+1]:null,defaultConfig:i})}else if(e===`ggml-stt`)if(t>=0){let e=process.argv[t+1];e||(console.error(`Error: --test-models requires a comma-separated list of model IDs`),process.exit(1)),await ra({modelIds:e.split(`,`).map(e=>e.trim()),defaultConfig:i})}else if(n)await ra({modelIds:s,defaultConfig:i});else{let e=process.argv.findIndex(e=>e===`--test-caps-model-id`);await oa({modelId:e>=0?process.argv[e+1]:null,defaultConfig:i})}else if(t>=0){let e=process.argv[t+1];e||(console.error(`Error: --test-models requires a comma-separated list of model IDs`),process.exit(1)),await ta({modelIds:e.split(`,`).map(e=>e.trim()),defaultConfig:i})}else if(n)await ta({modelIds:o,defaultConfig:i});else{let e=process.argv.findIndex(e=>e===`--test-caps-model-id`);await na({modelId:e>=0?process.argv[e+1]:null,defaultConfig:i})}}let u=Ii(i);t&&(u.server.port=t),u.server.port||(u.server.port=2080),_a({config:u,enableOpenAICompat:process.env.ENABLE_OPENAI_COMPAT_ENDPOINT===`1`||u.global.openai_compat?.enabled===!0,enableAnthropicMessages:process.env.ENABLE_ANTHROPIC_MESSAGES_ENDPOINT===`1`||u.global.anthropic_messages?.enabled===!0}).then(async({port:e,openaiEnabled:t,anthropicMessagesEnabled:n,autoDiscover:r})=>{let o=$i();console.log(`Buttress server listening on port ${e}`),console.log(`--------------------------------`),await ma(),console.log();let s={"ggml-llm":`LLM (GGML)`,"ggml-stt":`STT (GGML)`,"mlx-llm":`LLM (MLX)`};console.log(`Current supported Generators:`);let c=new Set((u?.generators||[]).map(e=>e.type).filter(Boolean));if(c.size===0)console.log(`- LLM (GGML)`),console.log(`- STT (GGML)`);else for(let e of c)console.log(`- ${s[e]||e}`);console.log(),console.log("Please configure `Buttress (Remote Inference)` in the Generator to connect to this server."),console.log(),console.log(`- Use http://${o}:${e} to connect to this server via LAN.`),console.log(`- Visit http://${o}:${e}/status to see status via LAN.`),console.log(),t?(console.log(`OpenAI-compatible API [EXPERIMENTAL]:`),console.log(`- Base URL: http://${o}:${e}/oai-compat/v1`),console.log(`- Chat completions: POST http://${o}:${e}/oai-compat/v1/chat/completions`),console.log(`- Models: GET http://${o}:${e}/oai-compat/v1/models`),console.log()):(console.log(`OpenAI-compatible API [EXPERIMENTAL]: disabled`),console.log(`  Set [openai_compat] enabled = true in config to enable`),console.log()),n?(console.log(`Anthropic Messages API [EXPERIMENTAL]:`),console.log(`- Base URL: http://${o}:${e}/anthropic-messages`),console.log(`- Messages: POST http://${o}:${e}/anthropic-messages/v1/messages`),console.log(`- Count tokens: POST http://${o}:${e}/anthropic-messages/v1/messages/count_tokens`),console.log()):(console.log(`Anthropic Messages API [EXPERIMENTAL]: disabled`),console.log(`  Set [anthropic_messages] enabled = true in config to enable`),console.log()),r&&(console.log(`Auto-discover enabled`),console.log()),i&&await a(i)}).catch(e=>{console.error(`Failed to start Buttress server:`,e),process.exitCode=1})};const{version:la,name:ua}=Zr(),da=async()=>{let e=`https://registry.npmjs.org/${ua}/latest`;try{let t=new AbortController,n=setTimeout(()=>t.abort(),3e3),r=await fetch(e,{headers:{Accept:`application/json`},signal:t.signal});return clearTimeout(n),r.ok&&(await r.json()).version||null}catch{return null}},fa=(e,t)=>{if(!t)return!1;let n=e.split(/[.-]/),r=t.split(/[.-]/);for(let e=0;e<Math.max(n.length,r.length);e+=1){let t=parseInt(n[e])||0,i=parseInt(r[e])||0;if(i>t)return!0;if(i<t)return!1}return!1},pa=e=>{console.log(``),console.log(`\x1B[33m╭─────────────────────────────────────────────────╮\x1B[0m`),console.log(`\x1B[33m│\x1B[0m  Update available! \x1B[2m%s\x1B[0m → \x1B[32m%s\x1B[0m`,la.padEnd(12),e.padEnd(12),`\x1B[33m│\x1B[0m`),console.log(`\x1B[33m│\x1B[0m                                                 \x1B[33m│\x1B[0m`),console.log(`\x1B[33m│\x1B[0m  Run to upgrade:                                \x1B[33m│\x1B[0m`),console.log(`\x1B[33m│\x1B[0m  \x1B[36mnpm install -g %s\x1B[0m      \x1B[33m│\x1B[0m`,ua.padEnd(27)),console.log(`\x1B[33m╰─────────────────────────────────────────────────╯\x1B[0m`),console.log(``)},ma=async()=>{try{let e=await da();e&&fa(la,e)&&pa(e)}catch{}},ha=typeof process<`u`&&process.versions&&process.versions.node,ga=async({backend:e,router:r,config:i,enableOpenAICompat:o,enableAnthropicMessages:l})=>{try{await c.mkdir(i.server.temp_file_dir,{recursive:!0})}catch{}let u=$i()||`0.0.0.0`,d={id:i.server.id,name:i.server.name,version:la,address:u,port:i.server.port,url:`http://${u}:${i.server.port}`,generators:Mi(i,i.generators.map(e=>e.type)),authentication:{required:!0,type:`device-group`}},f=new n({serve:{maxRequestBodySize:i.server.max_body_size},websocket:{idleTimeout:Math.ceil(i.server.session_timeout/1e3)},adapter:ha?t():void 0}).state({sessions:new Map,backend:e||Lr,config:i,serverInfo:d});r&&f.use(r),i.autodiscover?.http?.enabled&&f.use(ni(i)),f.use(ii),f.use(li),o&&f.use(vi(i)),l&&f.use(wi(i));let p={INVALID_REQUEST:-32600,INVALID_PARAMS:-32602,METHOD_NOT_FOUND:-32601,INTERNAL_ERROR:-32603},m=e=>e.id??e.raw?.id??e.remoteAddress;return f.ws(`/buttress/rpc`,{parse:(e,t)=>{if(typeof t==`string`)try{return JSON.parse(t)}catch{return e.send(JSON.stringify({jsonrpc:`2.0`,error:{code:p.INVALID_REQUEST,message:`Invalid request`},id:null})),null}return t},body:a.Object({jsonrpc:a.String(),method:a.String(),params:a.String(),id:a.String()}),open(e){let t=m(e);if(console.log(`[Request] New connection: ${t}`),!e.data.store.sessions.has(t))e.data.store.sessions.set(t,{streamReaders:new Map,generators:new Set,initializedContexts:new Set,timeout:null});else{let n=e.data.store.sessions.get(t);n&&(clearTimeout(n.timeout),n.timeout=null)}},async message(e,{id:t,method:n,params:r}){let i=m(e);console.log(`[Request] Received request from ${i}: ${n}`);let a=e.data.store.sessions.get(i),[o,c]=n.split(`.`),l=Ji[o]?.[c];if(!l){e.send(JSON.stringify({jsonrpc:`2.0`,error:{code:p.METHOD_NOT_FOUND,message:`Method not found`},id:t}));return}try{if(n===`cancel`){a.streamReaders.has(t)&&(a.streamReaders.get(t)?.cancel(),a.streamReaders.delete(t));return}if(n===`ping`){e.send(JSON.stringify({jsonrpc:`2.0`,result:`pong`,id:t}));return}let u=Yi(r),d=qi[o]?.[c],f=d?d.parse(u):u,p=await l({...e.data.store,peerId:i,session:a},...f);if(p instanceof s){let r=p.getReader();a.streamReaders.set(t,r),e.send(JSON.stringify({jsonrpc:`2.0`,result:{type:`stream`},id:t}));try{for(;;){let{value:n,done:i}=await r.read();if(i)break;let{event:a,data:o}=n;e.send(JSON.stringify({jsonrpc:`2.0`,method:`notification/${a}`,params:Xi(o),id:t}))}e.send(JSON.stringify({jsonrpc:`2.0`,method:`notification/_end`,id:t}))}catch(r){console.error(`[RPC] Stream error for ${n}:`,r),e.send(JSON.stringify({jsonrpc:`2.0`,method:`notification/_error`,params:Xi(r),id:t}))}a.streamReaders.delete(t)}else e.send(JSON.stringify({jsonrpc:`2.0`,result:Xi(p),id:t}))}catch(r){if(r instanceof ne){e.send(JSON.stringify({jsonrpc:`2.0`,error:{code:p.INVALID_PARAMS,message:`Invalid params`,data:r.issues},id:t}));return}console.error(`[RPC] Handler error for ${n}:`,r),e.send(JSON.stringify({jsonrpc:`2.0`,error:{code:p.INTERNAL_ERROR,message:String(r)},id:t}))}},async close(e){let t=m(e);console.log(`[Request] Connection closed: ${t}`);let{backend:n,sessions:r}=e.data.store,a=r.get(t);a&&(a.streamReaders.forEach(e=>e.cancel()),a.streamReaders.clear(),a.timeout=setTimeout(()=>{r.delete(t),console.log(`[Request] Session timed out: ${t}`),a.generators.forEach(e=>{n.finalizeGenerator(e)})},i.server.session_timeout))}}),{app:f,config:i}},_a=async({backend:e,router:t,config:n,enableOpenAICompat:r=!1,enableAnthropicMessages:i=!1})=>{let{app:a,config:o}=await ga({backend:e,router:t,config:n,enableOpenAICompat:r,enableAnthropicMessages:i}),{server:{port:s}}=o,c=[new Promise(e=>a.listen(s,e))],l=null;return o.autodiscover&&(l=new Qi(o.autodiscover,()=>a.store.serverInfo),c.push(l.start())),await Promise.all(c),{app:a,port:s,openaiEnabled:r,anthropicMessagesEnabled:i,autoDiscover:l}},va=[new URL(`index.mjs`,import.meta.url).pathname,new URL(`index.ts`,import.meta.url).pathname];(process.argv[1]?.endsWith(`/bricks-buttress`)||va.includes(process.argv[1]))&&await ca();export{ma as checkAndNotifyUpdates,da as checkForUpdates,fa as compareVersions,ga as createServer,pa as logUpdateMessage,Ii as processConfig,qr as startModelDownload,_a as startServer};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fugood/buttress-server",
-  "version": "2.24.2",
+  "version": "2.25.0-beta.6",
   "main": "lib/index.mjs",
   "types": "lib/index.d.mts",
   "type": "module",
@@ -44,5 +44,5 @@
     "tsdown": "^0.20.1",
     "typescript": "^5.9.3"
   },
-  "gitHead": "984a440ed04862f12c65f3cf62bdc70a938fcdd6"
+  "gitHead": "d8a3bbd695e9118cf725b7226866d804bbe56bbe"
 }

package/public/lib/index.d.ts

@@ -0,0 +1,27 @@
+import type { AnyElysia } from "elysia";
+import * as backendCore from "@fugood/buttress-backend-core";
+import { AutodiscoverService } from "./autodiscover";
+import type { Config } from "./types";
+export { startModelDownload } from "@fugood/buttress-backend-core";
+export { processConfig } from "./utils/config";
+export declare const checkForUpdates: () => Promise<string | null>;
+export declare const compareVersions: (current: string, latest: string) => boolean;
+export declare const logUpdateMessage: (latestVersion: string) => void;
+export declare const checkAndNotifyUpdates: () => Promise<void>;
+export type Backend = typeof backendCore;
+export interface StartServerOptions {
+	backend?: Backend;
+	router?: AnyElysia;
+	config: Config;
+	enableOpenAICompat?: boolean;
+}
+export declare const createServer: ({ backend, router, config, enableOpenAICompat }: StartServerOptions) => Promise<{
+	app: AnyElysia;
+	config: Config;
+}>;
+export declare const startServer: ({ backend, router, config, enableOpenAICompat }: StartServerOptions) => Promise<{
+	app: AnyElysia;
+	port: number;
+	openaiEnabled: boolean;
+	autoDiscover: AutodiscoverService | null;
+}>;

package/public/lib/index.mjs

@@ -0,0 +1,110 @@
+#!/usr/bin/env node
+import{createRequire as s3}from"node:module";var l3=Object.defineProperty;var i3=($,X)=>{for(var Z in X)l3($,Z,{get:X[Z],enumerable:!0,configurable:!0,set:(W)=>X[Z]=()=>W})};var P=($,X)=>()=>($&&(X=$($=0)),X);var R1=s3(import.meta.url);var w1=($,X,Z)=>Math.min(Math.max($,X),Z),F1=($)=>$?40:0,E1=($=0)=>{if(!$)return 0;return w1($/12884901888*20,0,20)},M1=($=0)=>{if(!$)return 0;return w1($/34359738368*10,0,10)},x1=($)=>$?10:0,S1=($="default",X=null)=>{let Z=String($).toLowerCase();if(!Z)return 0;if(Z.includes("cuda"))return 20;if(Z.includes("vulkan"))return 10;if(Z.includes("default"))return X==="darwin"||X==="ios"?15:5;return 0},q0=({platform:$,variant:X,hasGpu:Z,gpuUsableBytes:W=0,cpuUsableBytes:j=0,ok:V=!0}={})=>{if(!V)return 0;let Y=F1(Z)+S1(X,$)+E1(W),J=M1(j),N=x1(V);return Math.min(100,Math.round(Y+J+N))},F4=({platform:$,variant:X,hasGpu:Z,gpuUsableBytes:W=0,cpuUsableBytes:j=0,ok:V=!0}={})=>({gpuPresence:F1(Z),variant:S1(X,$),gpuMemory:E1(W),cpuMemory:M1(j),availability:x1(V)});var D1,u0=0.85,f0=0.5,P1=($)=>{if(!$&&$!==0)return[];if(Array.isArray($))return $.filter((X)=>X!=null);return[$]},n3=($)=>{if(!$)return null;return String($).trim().toLowerCase()||null},r3=({variant:$,preferVariants:X=[],variantPreference:Z=[],defaultVariants:W=D1}={})=>{let j=[];if($)j.push($);j.push(...P1(X)),j.push(...P1(Z)),j.push(...W);let V=j.map(n3).filter(Boolean);return Array.from(new Set(V))},T1=($={})=>{let X=String($.type||$.deviceType||$.kind||"").toLowerCase();if(X.includes("gpu"))return!0;if(X.includes("cuda"))return!0;if(X.includes("metal"))return!0;if(X.includes("vulkan"))return!0;if(X.includes("snapdragon"))return!0;return!1},o3=($)=>{if(!Array.isArray($))return[];return $.map((X)=>({...X}))},a3=($,X)=>{if($==="snapdragon")return X.filter((Z)=>Z.deviceName!=="GPUOpenCL");return X},k1=({platform:$,totalMemoryInBytes:X,variant:Z,devices:W,gpuMemoryFraction:j,cpuMemoryFraction:V,ok:Y,error:J})=>{let N=o3(a3(Z,W)),H=N.some(T1),O=N.filter((R)=>T1(R)&&Number.isFinite(Number(R.maxMemorySize))).reduce((R,U)=>R+U.maxMemorySize,0),q=X,_=H?Math.floor(O*j):0,G=q?Math.floor(q*V):0,w={platform:$,variant:Z,hasGpu:H,gpuUsableBytes:_,cpuUsableBytes:G,ok:Y},z=q0(w),A=Y?F4(w):null;return{platform:$,ok:Y,variant:Z,hasGpu:H,devices:N,gpuTotalBytes:O,gpuUsableBytes:_,cpuTotalBytes:q,cpuUsableBytes:G,score:z,breakdown:A,error:J,timestamp:new Date().toISOString()}},E4=({device:$,modelBytes:X=0,kvCacheBytes:Z=0}={})=>{if(!$)return{totalRequiredBytes:X+Z,fitsInGpu:!1,fitsInCpu:!1,limiting:"unknown-device"};let W=Math.max(0,Number(X)||0)+Math.max(0,Number(Z)||0),j=$.hasGpu&&W>0&&W<=$.gpuUsableBytes,V=W>0&&W<=$.cpuUsableBytes,Y="ok";if(!j&&$.hasGpu)Y="gpu-memory";if(!V)Y=j?"cpu-memory":"insufficient-memory";return{totalRequiredBytes:W,fitsInGpu:j,fitsInCpu:V,limiting:Y}},E0=async({platform:$,variant:X=null,preferVariants:Z=[],variantPreference:W=[],gpuMemoryFraction:j=u0,cpuMemoryFraction:V=f0,includeBreakdown:Y=!1,totalMemoryInBytes:J,modelBytes:N=null,kvCacheBytes:H=null,limitedKvCacheBytes:O=null,dependencies:q={},defaultVariants:_=D1}={})=>{let{getBackendDevicesInfo:G,isLibVariantAvailable:w}=q;if(typeof G!=="function"||typeof w!=="function")throw TypeError("GGML capability detection requires getBackendDevicesInfo and isLibVariantAvailable functions");let z=r3({variant:X,preferVariants:Z,variantPreference:W,defaultVariants:_}),A=[];for(let B of z)try{if(!await w(B))throw Error(`Variant ${B} not available on this platform`);let F=await G(B);A.push(k1({platform:$,totalMemoryInBytes:J,variant:B,devices:F,gpuMemoryFraction:j,cpuMemoryFraction:V,ok:!0}))}catch(L){let F=L instanceof Error?L.message:String(L);A.push(k1({platform:$,totalMemoryInBytes:J,variant:B,devices:[],gpuMemoryFraction:j,cpuMemoryFraction:V,ok:!1,error:F}))}let U=A.filter((B)=>B.ok)[0]||null,Q={ok:Boolean(U),selected:U?{...U,breakdown:Y?U.breakdown:void 0}:null,attempts:A};if(!Y&&Q.selected)delete Q.selected.breakdown;if(!Q||!N&&!H)return Q;let K=(B)=>{if(!B)return B;let L=E4({device:B,modelBytes:N||0,kvCacheBytes:H||0}),F=null;if(O!=null&&O!==H)F=E4({device:B,modelBytes:N||0,kvCacheBytes:O});return{...B,fit:L,...F&&{limitedFit:F}}};return Q.selected=K(Q.selected),Q.attempts=Array.isArray(Q.attempts)?Q.attempts.map(K):Q.attempts,Q},g0="ggml-llm";var p0=P(()=>{D1=["cuda","vulkan","snapdragon","default"]});var M4="ggml-stt",b1,x4=async({platform:$,variant:X=null,preferVariants:Z=[],variantPreference:W=[],gpuMemoryFraction:j=u0,cpuMemoryFraction:V=f0,includeBreakdown:Y=!1,totalMemoryInBytes:J,modelBytes:N=null,processingBytes:H=null,kvCacheBytes:O=null,dependencies:q={}}={})=>{let _=W&&W.length>0?W:b1;return E0({platform:$,variant:X,preferVariants:Z,variantPreference:_,gpuMemoryFraction:j,cpuMemoryFraction:V,includeBreakdown:Y,totalMemoryInBytes:J,modelBytes:N,kvCacheBytes:H??O,dependencies:q,defaultVariants:b1})};var S4=P(()=>{p0();b1=["cuda","vulkan","default"]});var t3,_0=async({platform:$,totalMemoryInBytes:X,backend:Z=g0,dependencies:W,...j}={})=>{let V=t3.get(Z);if(!V)throw Error(`No capability detector registered for backend "${Z}"`);return await V({...j,dependencies:W,totalMemoryInBytes:X,platform:$})};var v1=P(()=>{p0();S4();t3=new Map([[g0,E0],[M4,x4]])});var h1,P4=($)=>{let X=$?String($).toLowerCase():"f16";return h1[X]||h1.f16},T4=($,X,Z,W,j,V={},{totalLayers:Y=null,swaLayers:J=0,swaContext:N=null,swaContextMultiplier:H=1,swaAdditionalTokens:O=0,swaFull:q=!1}={})=>{if(!$||!X||!Z||!W||!j)return 0;let _=Y!=null&&Y!==void 0?Number(Y):Number($),G=Math.max(0,Math.floor(_));if(!G)return 0;let w=P4(V.k),z=P4(V.v),A=Number(Z)*(Number(W)*w+Number(j)*z);if(!A)return 0;let R=Math.max(0,Number(X)||0),U=Math.min(G,Math.max(0,Math.floor(Number(J)||0))),Q=Math.max(0,G-U),K=N!=null&&Number.isFinite(Number(N))?Math.max(0,Number(N)):R,B=Math.max(1,Number(H)||1),L=Math.max(0,Number(O)||0),F=K*B+L,E=q?R:Math.min(R,F),M=Q*R+U*Math.max(0,Math.floor(E));return Math.round(A*M)},m0=({modelBytes:$=0,audioLengthSeconds:X=30,sampleRate:Z=16000,bytesPerSample:W=4}={})=>{let j=Math.max(0,Number($)||0),V=Math.max(0,Math.floor(Math.max(0,X)*Z*W)),Y=1048576,J=1073741824,N;if(j<209715200)N=125829120;else if(j<524288000)N=146800640;else if(j<2147483648)N=157286400;else N=167772160;let H;if(j<209715200)H=73400320;else if(j<524288000)H=141557760;else if(j<2147483648)H=230686720;else H=230686720;let O;if(j<104857600)O=20971520;else if(j<209715200)O=31457280;else if(j<524288000)O=89128960;else if(j<2147483648)O=225443840;else O=377487360;let q=N+H+O,_=j+q+V;return{modelBytes:j,audioBufferBytes:V,processingBufferBytes:q,totalBytes:_}};var k4=P(()=>{h1={f16:2,f32:4,q8_0:1,q6_k:0.75,q5_k:0.625,q5_k_m:0.625,q5_k_s:0.625,q5_1:0.625,q5_0:0.625,q4_k:0.5,q4_k_m:0.5,q4_k_s:0.5,q4_1:0.5,q4_0:0.5,iq4_nl:0.5}});var D4=($)=>$?String($).trim().toLowerCase():null,e3=($={},X=null)=>{if(!$)return null;let Z=D4(X),W=Z?`${Z}.attention.sliding_window`:null,j=(W&&$[W]!=null?$[W]:null)??$["llama.attention.sliding_window"];if(j==null)return null;let V=Number(j);return Number.isFinite(V)?V:null},C1=($=0,X=0,Z=!1)=>{let W=Math.max(0,Math.floor(Number($)||0)),j=Math.max(0,Math.floor(Number(X)||0));if(!W||j===1)return 0;if(j<=0)return W;let V=Math.max(0,j-1),Y=Math.floor(W/j),J=W%j,N=Z?Math.max(0,J-1):Math.min(J,V);return Y*V+N},I1=({arch:$,nLayer:X=0})=>({arch:D4($),enabled:!1,window:null,pattern:null,denseFirst:!1,type:null,kvLayers:Math.max(0,Math.floor(Number(X)||0)),swaLayers:0}),$6,c0=({arch:$,metadata:X={},nLayer:Z=0}={})=>{let W=D4($||X["general.architecture"]),j=Math.max(0,Math.floor(Number(Z)||0)),V=e3(X,W),Y=W?$6.get(W):null;if(!Y)return I1({arch:W,nLayer:Z});let J=Y({nLayer:j,nSwa:V,metadata:X});if(!J||!J.enabled||!J.window||J.window<=0)return I1({arch:W,nLayer:Z});let N=Math.max(0,Math.floor(Number(J.pattern)||0)),H=J.kvLayers!=null&&Number.isFinite(Number(J.kvLayers))?Number(J.kvLayers):j,O=Math.max(0,Math.floor(H)),q=C1(O,N,Boolean(J.denseFirst));return{arch:W,enabled:q>0,window:J.window,pattern:N,denseFirst:Boolean(J.denseFirst),type:J.type||"standard",kvLayers:O,swaLayers:q}};var y1=P(()=>{$6=new Map([["llama4",({nSwa:$})=>{if($===0)return{enabled:!1};return{enabled:!0,window:$&&$>0?$:8192,pattern:4,type:"chunked"}}],["afmoe",({nSwa:$})=>{if(!$||$<=0)return{enabled:!1};return{enabled:!0,window:$,pattern:4,type:"standard"}}],["phi3",()=>({enabled:!1})],["gemma2",({nSwa:$})=>{let X=$&&$>0?$:4096;if(!X)return{enabled:!1};return{enabled:!0,window:X,pattern:2,type:"standard"}}],["gemma3",({nSwa:$})=>{if(!$||$<=0)return{enabled:!1};return{enabled:!0,window:$,pattern:6,type:"standard"}}],["gemma3n",({nLayer:$,nSwa:X})=>{if(!X||X<=0)return{enabled:!1};return{enabled:!0,window:X,pattern:5,type:"standard",kvLayers:Math.min(20,$)}}],["gemma-embedding",({nSwa:$})=>{if(!$||$<=0)return{enabled:!1};return{enabled:!0,window:$,pattern:6,type:"symmetric"}}],["cohere2",({nSwa:$})=>{if(!$||$<=0)return{enabled:!1};return{enabled:!0,window:$,pattern:4,type:"standard"}}],["olmo2",({nSwa:$})=>{if(!$||$<=0)return{enabled:!1};return{enabled:!0,window:$,pattern:4,type:"standard"}}],["exaone4",({nLayer:$,nSwa:X})=>{let Z=$>=64,W=null;if(X&&X>0)W=X;else if(Z)W=4096;if(!W)return{enabled:!1};return{enabled:!0,window:W,pattern:4,type:"standard"}}],["gpt-oss",({nSwa:$})=>{if(!$||$<=0)return{enabled:!1};return{enabled:!0,window:$,pattern:2,type:"standard"}}],["smallthinker",({nSwa:$})=>{if(!$||$<=0)return{enabled:!1};return{enabled:!0,window:4096,pattern:4,denseFirst:!0,type:"standard"}}]])});var X6,Z6,u1=($)=>$?String($).trim().toLowerCase():null,f1=($)=>{let X=u1($);return X?X6.has(X):!1},g1=($)=>{let X=u1($);return X?Z6.has(X):!1},b4=($)=>{if(f1($))return"recurrent";if(g1($))return"hybrid";return"transformer"},d0=($={})=>{let X=$["general.architecture"],Z=(F,E=null)=>{let M=$[F],x=Number(M);return Number.isFinite(x)?x:E},W=(F,E=null)=>{let M=$[F];if(Array.isArray(M))return M;let x=Number(M);return Number.isFinite(x)?x:E},j=X?Z(`${X}.context_length`,Z("llama.context_length")):null,V=X?Z(`${X}.block_count`,Z("llama.block_count")):null,Y=X?Z(`${X}.embedding_length`,Z("llama.embedding_length")):null,J=X?Z(`${X}.attention.head_count`,Z("llama.attention.head_count")):null,N=X?W(`${X}.attention.head_count_kv`,W("llama.attention.head_count_kv",J)):null,H=null,O=null;if(Array.isArray(N)){let F=N.filter((E)=>Number(E)>0);if(F.length>0)H=Math.max(...F.map(Number)),O=F.length;else H=0,O=0}else H=N;let q=X?Z(`${X}.attention.key_length`,Z("llama.attention.key_length")):null,_=X?Z(`${X}.attention.value_length`,Z("llama.attention.value_length")):null,G=$["general.quantization_version"]||null,w=$["general.file_type"]||null,z=X?Z(`${X}.ssm.conv_kernel`):null,A=X?Z(`${X}.ssm.state_size`):null,R=X?Z(`${X}.ssm.inner_size`):null,U=X?Z(`${X}.ssm.group_count`):null,Q=X?Z(`${X}.ssm.time_step_rank`):null,K=X?Z(`${X}.rwkv.head_size`):null,B=X?Z(`${X}.rwkv.token_shift_count`,2):null,L=O!=null&&V!=null?V-O:null;return{arch:X,nCtxTrain:j,nLayer:V,nEmbd:Y,nHead:J,nHeadKv:H,nEmbdHeadK:q,nEmbdHeadV:_,quantVersion:G,fileType:w,attentionLayerCount:O,recurrentLayerCount:L,ssmDConv:z,ssmDState:A,ssmDInner:R,ssmNGroup:U,ssmDtRank:Q,rwkvHeadSize:K,rwkvTokenShiftCount:B}},G0=({layerCount:$,headKvCount:X,embdHeadKCount:Z,embdHeadVCount:W,cacheTypes:j,swaConfig:V,kvUnified:Y=!1,nParallel:J=1,swaFull:N=!1,arch:H=null,attentionLayerCount:O=null})=>{let q=b4(H);if(q==="recurrent")return()=>0;let _=q==="hybrid"&&O!=null?Math.max(0,Math.floor(Number(O)||0)):$,G=V?.window&&Y?Math.max(1,Number(J)||1):1,w=Y?1:Math.max(1,Number(J)||1);return(z)=>T4(_,z,X,Z,W,j,{totalLayers:_,swaLayers:V?.swaLayers||0,swaContext:V?.window,swaFull:N,swaContextMultiplier:G})*w},l0=({nLayer:$,nEmbd:X,recurrentLayerCount:Z=null,nSeqMax:W=1,ssmDConv:j=null,ssmDState:V=null,ssmDInner:Y=null,ssmNGroup:J=null,ssmDtRank:N=null,rwkvHeadSize:H=null,rwkvTokenShiftCount:O=2,arch:q=null})=>{if(b4(q)==="transformer")return 0;let G=Z!=null?Math.max(0,Math.floor(Number(Z)||0)):Math.max(0,Math.floor(Number($)||0));if(G===0)return 0;let w=Math.max(1,Math.floor(Number(W)||1)),z=0,A=0;if(H!=null&&H>0&&X!=null&&X>0)z=Math.max(1,Number(O)||2)*X,A=X*H;else if(V!=null&&Y!=null){let Q=Math.max(0,Number(j)||0),K=Math.max(0,Number(V)||0),B=Math.max(0,Number(Y)||0),L=Math.max(1,Number(J)||1);if(Math.max(0,Number(N)||0)>0)z=Q>0?(Q-1)*2*L*K:0,A=Math.floor(K*B/2);else z=Q>0?(Q-1)*(B+2*L*K):0,A=K*B}else return 0;let R=4,U=(z+A)*w*G*R;return Math.max(0,U)},i0=({maxCtx:$,availableMemory:X,modelBytes:Z,kvBytesForCtx:W})=>{let j=Math.max(1,Math.floor(Number($)||0));if(!W||X<=Z)return j;let V=1,Y=j,J=j;while(V<=Y){let N=Math.floor((V+Y)/2);if(Z+W(N)<=X)J=N,V=N+1;else Y=N-1}return J};var p1=P(()=>{k4();X6=new Set(["mamba","mamba2","rwkv6","rwkv6qwen2","rwkv7","arwkv7"]),Z6=new Set(["jamba","falcon-h1","plamo2","granitehybrid","lfm2","lfm2moe","nemotron_h","nemotron_h_moe","qwen3next"])});var M0=P(()=>{v1();k4();p0();S4();y1();p1()});import{EventEmitter as W6}from"node:events";class h4{constructor($=j6){this.maxEntries=$,this.modelLoads=[],this.completions=[],this.transcriptions=[]}addModelLoad($){v4(this.modelLoads,$,this.maxEntries),e.emit("status:modelLoad",$),e.emit("status:change",{type:"modelLoad",entry:$})}addCompletion($){v4(this.completions,$,this.maxEntries),e.emit("status:completion",$),e.emit("status:change",{type:"completion",entry:$})}addTranscription($){v4(this.transcriptions,$,this.maxEntries),e.emit("status:transcription",$),e.emit("status:change",{type:"transcription",entry:$})}getModelLoadHistory(){return[...this.modelLoads].reverse()}getCompletionHistory(){return[...this.completions].reverse()}getTranscriptionHistory(){return[...this.transcriptions].reverse()}clear(){this.modelLoads=[],this.completions=[],this.transcriptions=[]}}function I4($){let X=(Z)=>$(Z);return e.on("status:change",X),()=>e.off("status:change",X)}function c1($){m1+=1;let X=m1,Z=I4($);return{subscriberId:X,unsubscribe:Z}}function C4($){let X=[];return{generators:Array.from($.entries()).filter(([,W])=>W.type==="ggml-llm").map(([W,j])=>{let{instance:V}=j,Y=[];if(V.contexts)Y=Array.from(V.contexts.entries()).map(([J,N])=>{let H={key:J,refCount:N.refCount,hasModel:Boolean(N.context)},O=N.context.parallel.getStatus();return H.parallelStatus=O,X.push({generatorId:W,contextKey:J,...O}),H});return{id:W,type:j.type,refCount:j.refCount,repoId:V.info?.model?.repoId||null,quantization:V.info?.model?.quantization||null,variant:V.info?.runtime?.variant||null,nCtx:V.info?.runtime?.n_ctx||null,nParallel:V.info?.runtime?.n_parallel||null,contexts:Y}}),parallelStatuses:X,history:{modelLoads:$0.getModelLoadHistory(),completions:$0.getCompletionHistory()}}}function y4($){return{generators:Array.from($.entries()).filter(([,Z])=>Z.type==="ggml-stt").map(([Z,W])=>{let{instance:j}=W,V=j.getStatus?.()||{},Y=V.queueStatus||{processing:!1,queuedCount:0};return{id:Z,type:W.type,refCount:W.refCount,repoId:j.info?.model?.repoId||null,quantization:j.info?.model?.quantization||null,modelType:j.info?.model?.modelType||null,variant:j.info?.runtime?.variant||null,hasContext:V.hasContext||!1,contextRefCount:V.contextRefCount||0,queueStatus:Y}}),history:{modelLoads:N0.getModelLoadHistory(),transcriptions:N0.getTranscriptionHistory()}}}function d1($){return{timestamp:new Date().toISOString(),ggmlLlm:C4($),ggmlStt:y4($)}}var j6=9999,e,v4=($,X,Z)=>{if($.push({...X,timestamp:X.timestamp||new Date().toISOString()}),$.length>Z)$.shift()},$0,N0,m1=0;var s0=P(()=>{e=new W6;e.setMaxListeners(100);$0=new h4,N0=new h4});import y from"node:path";import T0 from"node:os";import{stat as B0,mkdir as Y6,open as Q6,unlink as o,readFile as t1,writeFile as e1,rename as f4,readdir as J6}from"node:fs/promises";import{createHash as p4}from"node:crypto";import{gguf as V6}from"@huggingface/gguf";import{loadModel as N6,getBackendDevicesInfo as $2,isLibVariantAvailable as X2}from"@fugood/llama.node";import U6 from"bytes";import*as G6 from"node:stream/web";class L0{constructor($,X){this.config=$,this.plan=X,this.baseDir=$.runtime.cache_dir,this.enabled=$.runtime.session_cache?.enabled!==!1,this.maxSizeBytes=x6($.runtime.session_cache?.max_size_bytes,10737418240),this.maxEntries=$.runtime.session_cache?.max_entries||1000,this.metadata={variant:X.info?.runtime?.variant||null,n_gpu_layers:X.info?.runtime?.n_gpu_layers||0,n_ctx:X.info?.runtime?.n_ctx||0,modelPath:X.localPath,cacheTypeK:X.info?.runtime?.cache_type_k||"f16",cacheTypeV:X.info?.runtime?.cache_type_v||"f16",kvUnified:X.info?.runtime?.kv_unified??null,swaFull:X.info?.runtime?.swa_full??null,flashAttnType:X.info?.runtime?.flash_attn_type||"off",isRecurrent:!1,isHybrid:!1},this.cacheMap=null,this.initialized=!1}updateModelInfo($){if(!$)return;if(this.metadata.isRecurrent=Boolean($.is_recurrent),this.metadata.isHybrid=Boolean($.is_hybrid),this.metadata.isRecurrent||this.metadata.isHybrid)console.log(`[SessionCache] Model architecture: recurrent=${this.metadata.isRecurrent}, hybrid=${this.metadata.isHybrid}`)}requiresExactMatch(){return this.metadata.isRecurrent||this.metadata.isHybrid}static checkTokenPrefixMatch($,X){if($.length>X.length)return!1;for(let Z=0;Z<$.length;Z+=1)if($[Z]!==X[Z])return!1;return!0}static async tokenizeToArray($,X){let Z=await $.tokenize(X);return Array.from(Z?.tokens||[])}async findFormattedMatchForRecurrent($,X,Z){let W=await L0.tokenizeToArray(Z,X),j=$.map(async(_)=>{try{let G=await L0.tokenizeToArray(Z,_.fullText);if(L0.checkTokenPrefixMatch(G,W))return{entry:_,usePromptState:!1,tokenCount:G.length};if(_.promptStatePath&&_.promptText){let w=await L0.tokenizeToArray(Z,_.promptText);if(L0.checkTokenPrefixMatch(w,W))return{entry:_,usePromptState:!0,tokenCount:w.length}}return null}catch(G){return console.warn(`[SessionCache] Failed to check entry ${_.id}: ${G.message}`),null}}),Y=(await Promise.all(j)).find((_)=>_!==null);if(!Y)return console.log("[SessionCache] No token prefix match found for recurrent/hybrid model"),null;let{entry:J,usePromptState:N,tokenCount:H}=Y;console.log(`[SessionCache] Token prefix match: ${J.id} (${H} tokens, usePromptState=${N})`);let O=N?J.promptStatePath:J.stateFilePath;if(!await n1(O))return await this.removeStaleEntry(J),null;return J.lastAccessedAt=new Date().toISOString(),await x0(this.cacheMap,this.baseDir).catch(()=>{}),{entry:J,usePromptState:N}}async initialize(){if(!this.enabled||this.initialized)return;try{await H0($4(this.baseDir)),await H0(o0(this.baseDir)),await H0(U2(this.baseDir)),this.cacheMap=await A6(this.baseDir),this.initialized=!0,console.log(`[SessionCache] Initialized with ${Object.keys(this.cacheMap.entries).length} entries`)}catch($){console.warn(`[SessionCache] Failed to initialize: ${$.message}`),this.enabled=!1}}async removeStaleEntry($){if(console.log(`[SessionCache] Removing stale entry: ${$.id}`),$.stateFilePath)await o($.stateFilePath).catch(()=>{});if($.promptStatePath)await o($.promptStatePath).catch(()=>{});delete this.cacheMap.entries[$.id],this.cacheMap.totalSize-=($.stateFileSize||0)+($.promptStateSize||0),await x0(this.cacheMap,this.baseDir).catch(()=>{})}async findMatchingEntry($,X=null){if(!this.enabled||!this.cacheMap)return null;let Z=this.requiresExactMatch();if(Z&&X){let J=Object.values(this.cacheMap.entries).filter((N)=>m4(N.metadata,this.metadata)&&N.fullText);return this.findFormattedMatchForRecurrent(J,$,X)}let W=w6($,this.metadata,this.cacheMap,Z);if(!W)return null;let{entry:j}=W;if(!await n1(j.stateFilePath))return await this.removeStaleEntry(j),null;return j.lastAccessedAt=new Date().toISOString(),await x0(this.cacheMap,this.baseDir).catch(()=>{}),{entry:j,usePromptState:!1}}async prepareCompletionOptions($,X,Z=null){let W={options:$,cacheEntry:null,promptPrefix:null};if(!this.enabled)return W;let j=await this.findMatchingEntry(X,Z);if(!j)return W;let{entry:V,usePromptState:Y}=j,J=Y?V.promptStatePath:V.stateFilePath,N=Y?V.promptText:V.fullText;return console.log(`[SessionCache] Found matching entry: ${V.id} (${N.length} chars, usePromptState=${Y})`),{options:{...$,load_state_path:J},cacheEntry:V,promptPrefix:N}}async saveCompletionState($,X,Z,W=0,j=null){if(!this.enabled)return null;let V=$+X,Y=z6(V,this.metadata),J=()=>{if(Z)o(Z).catch(()=>{});if(j)o(j).catch(()=>{})};if(this.cacheMap.entries[Y]){console.log(`[SessionCache] Entry already exists for prompt: ${Y}, updating position`);let O=this.cacheMap.entries[Y];return O.lastAccessedAt=new Date().toISOString(),delete this.cacheMap.entries[Y],this.cacheMap.entries[Y]=O,await x0(this.cacheMap,this.baseDir).catch(()=>{}),J(),O}let N=s1(Y,this.baseDir),H=j?s1(`${Y}-prompt`,this.baseDir):null;try{await H0(y.dirname(N)),await f4(Z,N);let O=await B0(N),q=0;if(j&&H)try{await f4(j,H),q=(await B0(H)).size,console.log(`[SessionCache] Saved prompt state: ${H}`)}catch(G){console.warn(`[SessionCache] Failed to save prompt state: ${G.message}`)}let _={id:Y,promptText:$,completionText:X,fullText:V,promptTokenCount:W,stateFilePath:N,stateFileSize:O.size,promptStatePath:H||null,promptStateSize:q,metadata:{...this.metadata},createdAt:new Date().toISOString(),lastAccessedAt:new Date().toISOString()};if(this.cacheMap.entries[Y]=_,this.cacheMap.totalSize+=O.size+q,!this.requiresExactMatch())await E6(this.cacheMap,$,Y,this.metadata);return await F6(this.cacheMap,this.maxSizeBytes,this.maxEntries),await x0(this.cacheMap,this.baseDir),console.log(`[SessionCache] Saved entry: ${Y} (${O.size} bytes, ${V.length} chars)`),_}catch(O){return console.warn(`[SessionCache] Failed to save state: ${O.message}`),J(),null}}async generateTempStatePath(){return await H0(o0(this.baseDir)),B6(this.baseDir)}async cleanup(){await M6(this.baseDir)}}async function K2($,X,Z={}){let{globalDownloadManager:W=null}=Z,j=e0(X),V=await S6(j),Y=new L0(j,V);await Y.initialize();let J={id:$,type:"ggml-llm",config:j,plan:V,info:V.info,contexts:new Map,downloads:new Map,globalDownloadManager:W,sessionCache:Y,finalized:!1},N=async()=>{if(J.finalized)return;J.finalized=!0;let Q=Array.from(J.contexts.values()),K=Q.map((L)=>{if(L.released)return Promise.resolve(!1);if(L.releaseRequested||L.releaseTimer)return Promise.resolve(!1);if(L.refCount=Math.max(0,L.refCount-1),L.refCount>0)return Promise.resolve(!1);return r0(J,L)});if(await Promise.allSettled(K),Q.length===0||Q.every((L)=>L.released))await J.sessionCache.cleanup()},H=async(Q={})=>{let{onProgress:K}=Q,B=await D6(J,K);return J.sessionCache.updateModelInfo(B.modelInfo),{modelInfo:B.modelInfo?{...B.modelInfo}:null,runtime:{...J.plan.info.runtime},download:{...J.plan.info.download}}},O=async()=>{if(J.finalized)return!1;let Q=z0(J),K=J.contexts.get(Q);if(!K)return!1;return b6(J,K,!1)},q=async(Q={})=>{let{options:K={},useCache:B=!0}=Q,L=z0(J),F=J.contexts.get(L);if(!F)throw Error(`Context "${L}" not initialized`);await F.ready;let E=K.prompt||"",M=null,x=null;if(!E&&K.messages){({messages:M}=K),x={chatTemplate:K.chat_template||K.chatTemplate,jinja:K.jinja??!0,tools:K.tools,parallel_tool_calls:K.parallel_tool_calls,tool_choice:K.tool_choice,reasoning_format:K.reasoning_format,enable_thinking:K.enable_thinking,add_generation_prompt:K.add_generation_prompt,now:K.now,chat_template_kwargs:K.chat_template_kwargs};let I=await F.context.getFormattedChat(M,x.chatTemplate,x);E=I?.prompt||I||""}if(B&&J.sessionCache.enabled&&E){let{options:I}=await J.sessionCache.prepareCompletionOptions(K,E,F.context),b=await J.sessionCache.generateTempStatePath(),r=(await F.context.tokenize(E))?.tokens?.length||0,k={...I,save_state_path:b},d=J.sessionCache.requiresExactMatch(),v=Boolean(k.load_state_path),S=null;if(d&&!v)S=await J.sessionCache.generateTempStatePath(),k.save_prompt_state_path=S;let C={repoId:J.plan.info.model?.repoId||null,quantization:J.plan.info.model?.quantization||null,variant:J.plan.info.runtime?.variant||null};return T6(F.context,k,J.sessionCache,E,b,r,J.id,C,S)}let D={repoId:J.plan.info.model?.repoId||null,quantization:J.plan.info.model?.quantization||null,variant:J.plan.info.runtime?.variant||null};return P6(F.context,K,J.id,D)},_=async(Q={})=>{let{text:K="",params:B={}}=Q,L=z0(J),F=J.contexts.get(L);if(!F)throw Error(`Context "${L}" not initialized`);await F.ready;let E=await F.context.tokenize(K,B);if(!E)return{tokens:[]};let M=Array.from(E.tokens??[]).map((x)=>Number(x));return{...E,tokens:M}},G=async(Q={})=>{let{tokens:K=[]}=Q,B=z0(J),L=J.contexts.get(B);if(!L)throw Error(`Context "${B}" not initialized`);await L.ready;let F=K.map((E)=>Number(E));return L.context.detokenize(F)},w=async(Q={})=>{let{messages:K=[],template:B,params:L}=Q,F=z0(J),E=J.contexts.get(F);if(!E)throw Error(`Context "${F}" not initialized`);return await E.ready,await E.context.getFormattedChat(K,B,L)},z=()=>Array.from(J.contexts.values()).some((Q)=>!Q.released&&(Q.releaseRequested||Q.releaseTimer||Q.refCount>0)),A=()=>{J.finalized=!1},R=()=>{let Q=[],K=Array.from(J.contexts.entries()).map(([B,L])=>{let F={key:B,refCount:L.refCount,hasModel:Boolean(L.context)},E=L.context.parallel.getStatus();return F.parallelStatus=E,Q.push({contextKey:B,...E}),F});return{id:J.id,type:J.type,repoId:J.plan.info.model?.repoId||null,quantization:J.plan.info.model?.quantization||null,variant:J.plan.info.runtime?.variant||null,nCtx:J.plan.info.runtime?.n_ctx||null,nParallel:J.plan.info.runtime?.n_parallel||null,contexts:K,parallelStatuses:Q}},U=(Q)=>{let K=Array.from(J.contexts.entries()).map(([B,L])=>L.context.parallel.subscribeToStatus((F)=>{Q({contextKey:B,...F})}));return{remove:()=>{K.forEach((B)=>{if(B?.remove)B.remove()})}}};return{id:$,type:"ggml-llm",info:V.info,contexts:J.contexts,initContext:H,completion:q,tokenize:_,detokenize:G,applyChatTemplate:w,releaseContext:O,finalize:N,getStatus:R,subscribeParallelStatus:U,hasPendingReleases:z,resetFinalized:A}}async function q2($,X,Z={}){let{onProgress:W,onComplete:j,onError:V}=Z;try{let Y=e0($),J=await c4(Y),N=H2(Y,J),{repoId:H}=J;if(await P0(N,J.size)){if(console.log(`[Download] Model already exists: ${H} at ${N}`),typeof j==="function")j({localPath:N,repoId:H,alreadyExists:!0});return{started:!1,localPath:N,repoId:H,alreadyExists:!0}}let q=X.getDownload(N);if(q)return console.log(`[Download] Already downloading: ${H}`),q.then(()=>{if(typeof j==="function")j({localPath:N,repoId:H,joinedExisting:!0})}).catch((G)=>{if(typeof V==="function")V(G)}),{started:!1,localPath:N,repoId:H,alreadyDownloading:!0};console.log(`[Download] Starting download: ${H}`);let _=(async()=>{try{if(J.isSplit&&J.splitCount>0){let G=/-(\d{5})-of-(\d{5})\.gguf$/,w=y.dirname(N),z=J.splitCount,A=0;for(let R=1;R<=z;R+=1){let U=String(R).padStart(5,"0"),Q=J.filename.replace(G,`-${U}-of-${String(z).padStart(5,"0")}.gguf`),K=`${Y.model.base_url.replace(/\/+$/,"")}/${J.repoId}/resolve/${J.revision}/${Q}`,B=y.join(w,Q);if(!await P0(B))await a0(K,J.headers,B,null,(F)=>{if(F>=0&&Number.isFinite(F)){let E=(A+F)/z;if(console.log(`[Download] ${H}: ${Math.round(E*100)}%`),typeof W==="function")W(E)}});A+=1}}else await a0(J.url,J.headers,N,J.size,(G)=>{if(G>=0&&Number.isFinite(G)){if(console.log(`[Download] ${H}: ${Math.round(G*100)}%`),typeof W==="function")W(G)}});if(console.log(`[Download] Completed: ${H}`),typeof j==="function")j({localPath:N,repoId:H})}catch(G){if(console.error(`[Download] Failed: ${H}`,G.message),typeof V==="function")V(G);throw G}finally{X.deleteDownload(N)}})();return X.setDownload(N,_),{started:!0,localPath:N,repoId:H}}catch(Y){if(console.error("[Download] Failed to start download:",Y.message),typeof V==="function")V(Y);return{started:!1,localPath:null,repoId:null,error:Y.message}}}async function v6($){let X=e0($),Z=await c4(X),W=await G2(Z.url,Z.headers,X.runtime.cache_dir),{arch:j,nCtxTrain:V,nLayer:Y,nEmbd:J,nHead:N,nHeadKv:H,nEmbdHeadK:O,nEmbdHeadV:q,quantVersion:_,fileType:G,attentionLayerCount:w,recurrentLayerCount:z,ssmDConv:A,ssmDState:R,ssmDInner:U,ssmNGroup:Q,ssmDtRank:K,rwkvHeadSize:B,rwkvTokenShiftCount:L}=d0(W),F=Number.isFinite(Number(Y))?Number(Y):0,E=Number.isFinite(Number(J))?Number(J):0,M=Number.isFinite(Number(N))?Number(N):0,x=Number.isFinite(Number(H))?Number(H):M,D=M>0&&E>0?E/M:128,I=O!=null&&Number.isFinite(Number(O))?Number(O):D,b=q!=null&&Number.isFinite(Number(q))?Number(q):D,u=c0({arch:j,metadata:W,nLayer:F}),r=u&&Number.isFinite(Number(u.kvLayers))?Number(u.kvLayers):F,k=Math.max(0,Math.floor(Number(r)||0)),v=(X.model.n_ctx?Number(X.model.n_ctx):null)||V||4096,S={k:X.model.cache_type_k,v:X.model.cache_type_v},C=Z.size>0?Z.size:0,h=X.model.n_parallel||4,p=G0({layerCount:k,headKvCount:x,embdHeadKCount:I,embdHeadVCount:b,cacheTypes:S,swaConfig:u,kvUnified:X.model.kv_unified,nParallel:h,swaFull:X.model.swa_full,arch:j,attentionLayerCount:w}),f=l0({nLayer:F,nEmbd:E,recurrentLayerCount:z,nSeqMax:h,ssmDConv:A,ssmDState:R,ssmDInner:U,ssmNGroup:Q,ssmDtRank:K,rwkvHeadSize:B,rwkvTokenShiftCount:L,arch:j}),s=X.backend?.gpu_memory_fraction!=null?Math.min(1,Math.max(0,Number(X.backend.gpu_memory_fraction))):S0.backend.gpu_memory_fraction||1,W0=X.backend?.cpu_memory_fraction!=null?Math.min(1,Math.max(0,Number(X.backend.cpu_memory_fraction))):t0,A0=p(v),l=await O2(X,{modelBytes:C,kvCacheBytes:A0}),F0=(l.selected.totalMemory||0)*s,A4=Math.max(0,T0.totalmem()*W0),h0=l.selected.hasGpu?F0:A4,I0=i0({maxCtx:v,availableMemory:h0,modelBytes:C,kvBytesForCtx:p}),U0=p(v),C0=p(I0);return{kvInfo:{nCtxTrain:V,nLayer:F,nEmbd:E,nHeadKv:x,nEmbdHeadK:I,nEmbdHeadV:b,nHeadCount:M,nHeadKvCount:x,kvLayerCount:k,swa:u?.enabled?{window:u.window,pattern:u.pattern,denseFirst:u.denseFirst,type:u.type,layers:u.swaLayers}:null},modelBytes:C,kvCacheBytes:U0,limitedKvCacheBytes:C0,memoryLimitedCtx:I0,recurrentMemoryBytes:f,quantization:{name:Z.quantization||null,fileType:G,version:_}}}async function _2($=null,X={}){let{threshold:Z=1.1,includeBreakdown:W=!1,config:j,...V}=X,Y=null,J=null,N=null,H=null,O=null,q=null,_=null;if(j)try{let{modelBytes:B,kvCacheBytes:L,limitedKvCacheBytes:F,memoryLimitedCtx:E,recurrentMemoryBytes:M,kvInfo:x,quantization:D}=await v6(j);Y=B,J=L,N=F,H=E,O=M,q=x,_=D}catch(B){}let G=j?.backend?.gpu_memory_fraction!=null?Math.min(1,Math.max(0,Number(j.backend.gpu_memory_fraction))):void 0,w=j?.backend?.cpu_memory_fraction!=null?Math.min(1,Math.max(0,Number(j.backend.cpu_memory_fraction))):void 0,z=await _0({...V,platform:process.platform,totalMemoryInBytes:T0.totalmem(),backend:"ggml-llm",includeBreakdown:W,gpuMemoryFraction:G,cpuMemoryFraction:w,dependencies:{getBackendDevicesInfo:$2,isLibVariantAvailable:X2},modelBytes:Y,kvCacheBytes:J,limitedKvCacheBytes:N}),A=z.selected,R=a1(A);A.modelBytes=Y||null,A.kvCacheBytes=J||null,A.memoryLimitedCtx=H||null,A.limitedKvCacheBytes=N||null,A.recurrentMemoryBytes=O||null,A.kvInfo=q||null,A.quantization=_||null;let U=null,Q=null;if($){let B=a1($);Q={...$,score:B};let L="buttress",F="buttress-higher-score";if(!z.ok)L="local",F="buttress-unavailable";else if(!B&&B!==0)L="buttress",F="missing-client-score";else{let{fit:E,limitedFit:M}=Q,x=A?.fit,D=A?.limitedFit,I=E?.fitsInGpu||E?.fitsInCpu||M?.fitsInGpu||M?.fitsInCpu,b=x?.fitsInGpu||x?.fitsInCpu||D?.fitsInGpu||D?.fitsInCpu;if(I&&!b)L="local",F="client-fits-in-memory";else if(b&&!I)L="buttress",F="buttress-fits-in-memory";else if(B>R*Z)L="local",F="client-better";else if(R>B*Z)L="buttress",F="buttress-better";else L="either",F="comparable-scores"}U={buttressScore:R,clientScore:B,threshold:Z,recommendation:L,reason:F}}if(!z.ok&&!U)U={buttressScore:R,clientScore:$?.score??null,threshold:Z,recommendation:"local",reason:"buttress-unavailable"};let K=null;if(j)K={repoId:j.model?.repo_id||null,quantization:j.model?.quantization||null,nCtx:j.model?.n_ctx||null,cacheKType:j.model?.cache_type_k||"f16",cacheVType:j.model?.cache_type_v||"f16"};return{type:"ggml-llm",timestamp:new Date().toISOString(),buttress:z,client:Q,comparison:U,modelConfig:K}}var H6=()=>{if(typeof globalThis<"u"&&globalThis.ReadableStream&&globalThis.WritableStream)return{ReadableStream:globalThis.ReadableStream,WritableStream:globalThis.WritableStream};return G6},O6,Z2,K6,W2=($={},X={})=>{return Object.entries(X||{}).forEach(([Z,W])=>{if(W&&typeof W==="object"&&!Array.isArray(W)){if(!$[Z]||typeof $[Z]!=="object")$[Z]={};W2($[Z],W)}else $[Z]=W}),$},q6=".gguf",j2="https://huggingface.co",Y2="https://huggingface.co/api",n,Q2,t0=0.5,S0,u4=($,X=[])=>{if(!$&&$!==0)return[...X];if(Array.isArray($))return $.filter((Z)=>Z!=null);return[$]},n0=($)=>{if(!$)return null;let X=String($).toLowerCase();if(["cuda","vulkan","snapdragon","default"].includes(X))return X;return null},e0=($={})=>{let X=JSON.parse(JSON.stringify(S0));if(W2(X,$),X.backend.variant=n0(X.backend.variant),X.backend.variant_preference=Array.from(new Set(u4(X.backend.variant_preference).map(n0).filter(Boolean))),X.backend.variant_preference.length===0)X.backend.variant_preference=["cuda","vulkan","snapdragon","default"];if(X.runtime.prefer_variants=Array.from(new Set(u4(X.runtime.prefer_variants).map(n0).filter(Boolean))),X.model.preferred_quantizations=Array.from(new Set(u4(X.model.preferred_quantizations||X.model.quantizations).map((Z)=>Z?String(Z).toLowerCase():null).filter(Boolean))),X.model.quantization){let Z=String(X.model.quantization).toLowerCase();if(!X.model.preferred_quantizations.includes(Z))X.model.preferred_quantizations.unshift(Z)}return X.model.n_parallel=Math.max(1,Number(X.model.n_parallel)||4),X.model.n_batch=Math.max(1,Number(X.model.n_batch)||512),X.model.base_url=X.model.base_url||j2,X.model.api_base=X.model.api_base||Y2,X.runtime.cache_dir=X.runtime.cache_dir?y.resolve(X.runtime.cache_dir):n,X.runtime.session_cache={...S0.runtime.session_cache,...X.runtime.session_cache||{}},X.runtime.context_release_delay_ms=Math.max(0,Number(X.runtime.context_release_delay_ms)||S0.runtime.context_release_delay_ms),X},l1=($)=>{let X=$.toLowerCase();return Q2.find((W)=>X.includes(W))||null},_6=($)=>{let X=[];if($.backend.variant)X.push($.backend.variant);if($.runtime.prefer_variants.length>0)X.push(...$.runtime.prefer_variants);return X.push(...$.backend.variant_preference),X.push("default"),Array.from(new Set(X.map(n0).filter(Boolean)))},H0=async($)=>{await Y6($,{recursive:!0})},L6=($=n)=>y.join($,".metadata-cache"),J2=($,X,Z=n)=>{let W=p4("sha256").update($).digest("hex");return y.join(L6(Z),X,`${W}.json`)},V2=async($,X,Z=n)=>{try{let W=J2($,X,Z),j=await t1(W,"utf-8");return console.log(`[Cache] Hit ${X} cache:`,y.basename(W)),JSON.parse(j,(V,Y)=>{if(typeof Y==="string"&&Y.startsWith("__bigint__"))return BigInt(Y.slice(10));return Y})}catch(W){return null}},g4=async($,X,Z,W=n)=>{try{let j=J2($,X,W);await H0(y.dirname(j)),await e1(j,JSON.stringify(Z,(V,Y)=>{if(typeof Y==="bigint")return`__bigint__${Y.toString()}`;return Y}),"utf-8"),console.log(`[Cache] Wrote ${X} cache:`,y.basename(j))}catch(j){console.warn(`[Cache] Failed to write ${X} cache:`,j.message)}},$4=($=n)=>y.join($,".session-state-cache"),N2=($=n)=>y.join($4($),"cache-map.json"),o0=($=n)=>y.join($4($),"temp"),U2=($=n)=>y.join($4($),"states"),i1=()=>({version:1,entries:{},totalSize:0}),A6=async($=n)=>{try{let X=N2($),Z=await t1(X,"utf-8"),W=JSON.parse(Z);if(!W.entries||typeof W.entries!=="object")return i1();return W}catch{return i1()}},x0=async($,X=n)=>{let Z=N2(X),W=`${Z}.tmp.${Date.now()}`;try{await H0(y.dirname(Z)),await e1(W,JSON.stringify($,null,2),"utf-8"),await f4(W,Z)}catch(j){throw await o(W).catch(()=>{}),j}},z6=($,X)=>{let Z=JSON.stringify({text:$,model:X.modelPath,variant:X.variant,n_gpu_layers:X.n_gpu_layers,n_ctx:X.n_ctx,cacheTypeK:X.cacheTypeK,cacheTypeV:X.cacheTypeV,kvUnified:X.kvUnified,swaFull:X.swaFull,flashAttnType:X.flashAttnType});return p4("sha256").update(Z).digest("hex").slice(0,24)},s1=($,X=n)=>y.join(U2(X),`${$}.bin`),B6=($=n)=>{let X=`${Date.now()}-${Math.random().toString(36).slice(2,10)}`;return y.join(o0($),`${X}.bin`)},m4=($,X)=>$.modelPath===X.modelPath&&$.variant===X.variant&&$.n_gpu_layers===X.n_gpu_layers&&$.n_ctx>=X.n_ctx&&$.cacheTypeK===X.cacheTypeK&&$.cacheTypeV===X.cacheTypeV&&$.kvUnified===X.kvUnified&&$.swaFull===X.swaFull&&$.flashAttnType===X.flashAttnType&&Boolean($.isRecurrent)===Boolean(X.isRecurrent)&&Boolean($.isHybrid)===Boolean(X.isHybrid),R6=($,X)=>{let Z=Math.min($.length,X.length),W=0;while(W<Z&&$[W]===X[W])W+=1;return W},w6=($,X,Z,W=!1)=>{let j=Object.values(Z.entries);console.log(`[SessionCache] Finding match for promptText (${$.length} chars), exactMatch=${W}`),console.log(`[SessionCache] Checking ${j.length} cache entries`);let V=j.filter((J)=>m4(J.metadata,X));if(W){let J=V.find((N)=>N.fullText===$);if(J)return console.log(`[SessionCache] Exact match found: ${J.id} (${J.fullText.length} chars)`),{entry:J,prefixLength:J.fullText.length,exactMatch:!0};return null}let Y=V.reduce((J,N)=>{let H=R6($,N.fullText);if(H>J.prefixLen)return{entry:N,prefixLen:H};if(H===J.prefixLen&&N.fullText.length>(J.entry?.fullText?.length||0))return{entry:N,prefixLen:H};return J},{entry:null,prefixLen:0});if(Y.entry)return console.log(`[SessionCache] Prefix match found: ${Y.entry.id} (${Y.prefixLen}/${Y.entry.fullText.length} chars)`),{entry:Y.entry,prefixLength:Y.prefixLen};return console.log("[SessionCache] No match found"),null},F6=async($,X,Z)=>{let W=Object.values($.entries).sort((J,N)=>new Date(J.lastAccessedAt)-new Date(N.lastAccessedAt)),j=$.totalSize,V=Object.keys($.entries).length,Y=W.filter((J)=>{let N=j>X,H=V>Z;if(!N&&!H)return!1;return j-=(J.stateFileSize||0)+(J.promptStateSize||0),V-=1,!0});return await Promise.all(Y.map(async(J)=>{if(await o(J.stateFilePath).catch(()=>{}),J.promptStatePath)await o(J.promptStatePath).catch(()=>{});delete $.entries[J.id],console.log(`[SessionCache] Evicted entry: ${J.id}`)})),$.totalSize=Math.max(0,j),Y.map((J)=>J.id)},E6=async($,X,Z,W)=>{let j=Object.entries($.entries).filter(([V,Y])=>{if(V===Z)return!1;if(!m4(Y.metadata,W))return!1;return X.startsWith(Y.fullText)&&Y.fullText.length<X.length}).map(([,V])=>V);return await Promise.all(j.map(async(V)=>{if(await o(V.stateFilePath).catch(()=>{}),V.promptStatePath)await o(V.promptStatePath).catch(()=>{});$.totalSize-=(V.stateFileSize||0)+(V.promptStateSize||0),delete $.entries[V.id],console.log(`[SessionCache] Evicted superseded prefix entry: ${V.id} (${V.promptText.length} prompt chars)`)})),j.map((V)=>V.id)},M6=async($=n)=>{let X=o0($);try{let Z=await J6(X),W=Date.now(),j=3600000;await Promise.all(Z.map(async(V)=>{let Y=y.join(X,V),J=await B0(Y).catch(()=>null);if(J&&W-J.mtimeMs>3600000)await o(Y).catch(()=>{}),console.log(`[SessionCache] Cleaned up temp file: ${V}`)}))}catch{}},n1=async($)=>{try{return await B0($),!0}catch{return!1}},x6=($,X)=>{if($==null)return X;if(typeof $==="number")return $;if(typeof $==="string"){let Z=U6.parse($);return Z!=null?Z:X}return X},r1=async($,X={})=>{if(typeof fetch!=="function")throw Error("Global fetch is not available in this runtime");let Z=await fetch($,X);if(!Z.ok){let W=await Z.text().catch(()=>"");throw Error(`Failed to fetch ${$}: ${Z.status} ${Z.statusText} ${W}`.trim())}return Z.json()},o1=async($,X={})=>{if(typeof fetch!=="function")throw Error("Global fetch is not available in this runtime");let Z=await fetch($,{...X,method:"HEAD"});if(!Z.ok)throw Error(`Failed to fetch headers for ${$}: ${Z.status} ${Z.statusText}`);return Z},G2=async($,X,Z=n)=>{let W=JSON.stringify({url:$,headers:X}),j=await V2(W,"range-metadata",Z);if(j)return j;let V=!/^https?:/i.test($),{metadata:Y}=await V6($,{fetch,additionalFetchHeaders:X,allowLocalFile:V});return await g4(W,"range-metadata",Y,Z),Y},H2=($,X)=>{if($.model.local_path)return y.resolve($.model.local_path);let Z=X.repoId.split("/"),W=y.join($.runtime.cache_dir,...Z,X.revision);return y.join(W,X.filename)},P0=async($,X)=>{try{let Z=await B0($);if(!X)return!0;return Z.size===X}catch(Z){return!1}},a0=async($,X,Z,W,j)=>{if(typeof fetch!=="function")throw Error("Global fetch is not available in this runtime");await H0(y.dirname(Z));let V=await fetch($,{headers:X});if(!V.ok||!V.body)throw Error(`Failed to download ${$}: ${V.status} ${V.statusText}`);let Y=await Q6(Z,"w"),J=Number(V.headers.get("content-length"))||W||0,N=0,H=0.05;try{await V.body.pipeTo(new K6({async write(O){if(await Y.write(O),N+=O.byteLength,typeof j==="function"&&J>0){let q=Math.min(1,N/J);while(q>=H)j(H),H+=0.05}},async close(){if(await Y.close(),typeof j==="function")j(1)},async abort(O){throw await Y.close().catch(()=>{}),await o(Z).catch(()=>{}),O}}))}catch(O){throw await Y.close().catch(()=>{}),await o(Z).catch(()=>{}),O}if(W){let O=await B0(Z);if(O.size!==W)throw await o(Z).catch(()=>{}),Error(`Downloaded file size mismatch, expected ${W} got ${O.size}`)}},c4=async($)=>{let X=$.model.repo_id||$.model.repository||$.model.model;if(!X)throw Error("`model.repo_id` is required in Buttress backend config");let Z=$.model.revision||"main",W=$.runtime.cache_dir,j=JSON.stringify({repoId:X,revision:Z,filename:$.model.filename,url:$.model.url,quantization:$.model.quantization,preferred_quantizations:$.model.preferred_quantizations}),V=await V2(j,"artifact-info",W);if(V)return V;let Y={...$.runtime.http_headers||{}};if($.runtime.huggingface_token)Y.Authorization=`Bearer ${$.runtime.huggingface_token}`;if($.model.url){let Q=await o1($.model.url,{headers:Y}),K=Number(Q.headers.get("content-length"))||null,B=$.model.filename||$.model.url.split("/").pop(),L={repoId:X,revision:Z,filename:B,url:$.model.url,size:K,headers:Y};return await g4(j,"artifact-info",L,W),L}let{filename:J}=$.model,N=$.model.quantization&&String($.model.quantization).toLowerCase(),H=await r1(`${$.model.api_base}/models/${X}?revision=${Z}&blobs=true`,{headers:Y}),q=(H?.siblings||H?.files||[]).map((Q)=>Q.rfilename||Q.path||Q.filename).filter((Q)=>typeof Q==="string"&&Q.endsWith(q6));if(q.length===0)throw Error(`No GGUF artifacts found in repo ${X}`);let _=$.model.preferred_quantizations.length>0?$.model.preferred_quantizations:Q2,G=()=>{let Q=_.find((K)=>{return q.find((L)=>L.toLowerCase().includes(K))});if(Q)return{filename:q.find((B)=>B.toLowerCase().includes(Q)),quantization:Q};return null};if(!J){let Q=G()||{filename:q[0],quantization:null},{filename:K,quantization:B}=Q;J=K,N=B||l1(J)}else if(!N)N=l1(J);let w=`${$.model.base_url.replace(/\/+$/,"")}/${X}/resolve/${Z}/${J}`,z=/-(\d{5})-of-(\d{5})\.gguf$/,A=J.match(z),R=null;if(A){let[,,Q]=A,K=await r1(`${$.model.api_base}/models/${X}?revision=${Z}&blobs=true`,{headers:Y}),B=K?.siblings||K?.files||[],L=Number(Q);R=0;for(let F=1;F<=L;F+=1){let E=String(F).padStart(5,"0"),M=J.replace(z,`-${E}-of-${Q}.gguf`),x=B.find((I)=>(I.rfilename||I.path||I.filename)===M),D=Number(x?.size);if(Number.isFinite(D)&&D>0)R+=D}}else{let Q=await o1(w,{headers:Y});R=Number(Q.headers.get("content-length"))||null}let U={repoId:X,revision:Z,filename:J,url:w,size:R,quantization:N,headers:Y,isSplit:Boolean(A),splitCount:A?Number(A[2]):0};return await g4(j,"artifact-info",U,W),U},O2=async($,{modelBytes:X=null,kvCacheBytes:Z=null}={})=>{let W=_6($),[j,...V]=W,Y=$.backend?.gpu_memory_fraction!=null?Math.min(1,Math.max(0,Number($.backend.gpu_memory_fraction))):S0.backend.gpu_memory_fraction||1,J=$.backend?.cpu_memory_fraction!=null?Math.min(1,Math.max(0,Number($.backend.cpu_memory_fraction))):t0,N=await _0({platform:process.platform,totalMemoryInBytes:T0.totalmem(),backend:"ggml-llm",variant:j||null,preferVariants:V,gpuMemoryFraction:Y,cpuMemoryFraction:J,dependencies:{getBackendDevicesInfo:$2,isLibVariantAvailable:X2},modelBytes:X,kvCacheBytes:Z}),H=(_)=>({..._,devices:Array.isArray(_.devices)?_.devices:[],ok:_.ok,hasGpu:Boolean(_.hasGpu),totalMemory:_.gpuTotalBytes||_.totalMemory||0,error:_.ok?null:Error(_.error||`Variant ${_.variant} not available on this platform`)});if(!N.ok||!N.selected){let _=(N.attempts||[]).map((G)=>`${G.variant}: ${G.error||"unknown error"}`).join("; ");throw Error(`Unable to initialize any backend variant (${W.join(", ")}). Errors: ${_}`)}let O=(N.attempts||[]).map(H);return{selected:H(N.selected),attempts:O}},S6=async($)=>{let X=await c4($),Z=await G2(X.url,X.headers,$.runtime.cache_dir),{arch:W,nCtxTrain:j,nLayer:V,nEmbd:Y,nHead:J,nHeadKv:N,nEmbdHeadK:H,nEmbdHeadV:O,quantVersion:q,fileType:_,attentionLayerCount:G,recurrentLayerCount:w,ssmDConv:z,ssmDState:A,ssmDInner:R,ssmNGroup:U,ssmDtRank:Q,rwkvHeadSize:K,rwkvTokenShiftCount:B}=d0(Z),L=Number.isFinite(Number(V))?Number(V):0,F=Number.isFinite(Number(Y))?Number(Y):0,E=Number.isFinite(Number(J))?Number(J):0,M=Number.isFinite(Number(N))?Number(N):E,x=E>0&&F>0?F/E:128,D=H!=null&&Number.isFinite(Number(H))?Number(H):x,I=O!=null&&Number.isFinite(Number(O))?Number(O):x,b=c0({arch:W,metadata:Z,nLayer:L}),u=b&&Number.isFinite(Number(b.kvLayers))?Number(b.kvLayers):L,r=Math.max(0,Math.floor(Number(u)||0)),k={use_mmap:$.model.use_mmap??$.runtime.use_mmap,use_mlock:$.model.use_mlock??$.runtime.use_mlock,n_threads:$.model.n_threads??$.runtime.n_threads,n_ctx:$.model.n_ctx??$.runtime.n_ctx,n_batch:$.model.n_batch??$.runtime.n_batch,n_ubatch:$.model.n_ubatch??$.runtime.n_ubatch,n_cpu_moe:$.model.n_cpu_moe??$.runtime.n_cpu_moe,n_parallel:$.model.n_parallel??$.runtime.n_parallel,cpu_mask:$.model.cpu_mask??$.runtime.cpu_mask,cpu_strict:$.model.cpu_strict??$.runtime.cpu_strict,devices:$.model.devices??$.runtime.devices,n_gpu_layers:$.model.n_gpu_layers??$.runtime.n_gpu_layers,flash_attn_type:$.model.flash_attn_type??$.runtime.flash_attn_type,cache_type_k:$.model.cache_type_k??$.runtime.cache_type_k,cache_type_v:$.model.cache_type_v??$.runtime.cache_type_v,kv_unified:$.model.kv_unified??$.runtime.kv_unified,swa_full:$.model.swa_full??$.runtime.swa_full,ctx_shift:$.model.ctx_shift??$.runtime.ctx_shift},d=k.n_ctx?Number(k.n_ctx):null,v=d||j||4096,S=[],C=[],h=!0;if(d&&j&&d>j){h=!1;let V0=`Requested context length (${d}) exceeds model training context (${j})`;S.push(V0),C.push(V0),v=j}if(d&&!j)S.push("Model metadata missing training context length, using requested value");let p={k:k.cache_type_k,v:k.cache_type_v},f=X.size>0?X.size:0,s=G0({layerCount:r,headKvCount:M,embdHeadKCount:D,embdHeadVCount:I,cacheTypes:p,swaConfig:b,kvUnified:k.kv_unified,nParallel:k.n_parallel,swaFull:k.swa_full,arch:W,attentionLayerCount:G}),W0=l0({nLayer:L,nEmbd:F,recurrentLayerCount:w,nSeqMax:k.n_parallel||4,ssmDConv:z,ssmDState:A,ssmDInner:R,ssmNGroup:U,ssmDtRank:Q,rwkvHeadSize:K,rwkvTokenShiftCount:B,arch:W}),A0=s(v),l=await O2($,{modelBytes:f,kvCacheBytes:A0+W0}),L4=l.selected.totalMemory||0,F0=L4*($.backend.gpu_memory_fraction||1),A4=$.backend.cpu_memory_fraction!=null?Math.min(1,Math.max(0,Number($.backend.cpu_memory_fraction))):t0,h0=Math.max(0,T0.totalmem()*A4),I0=l.selected.hasGpu?F0:h0,U0=i0({maxCtx:v,availableMemory:I0,modelBytes:f,kvBytesForCtx:s});if(!d&&U0){let V0=j?Math.min(U0,j):U0,R4=Math.max(32,V0);if(R4<v)S.push(`Context length capped to ${R4} by memory limits`);v=R4}if(v>U0)v=U0;let C0=Math.floor(U0);console.log(`[buttress] Memory-limited context length: ${C0}`);let A1=s(v),p3=f+A1+W0,z1=L?f/(L+1):f,y0=0;if(l.selected.hasGpu&&z1>0)y0=Math.min(L+1,Math.max(0,Math.floor(F0/z1)));console.log(`[buttress] Auto GPU layer capacity (${l.selected.variant}): ${y0}/${L+1}`);let z4;if(k.n_gpu_layers==="auto"||k.n_gpu_layers==null)z4=y0;else z4=Math.max(0,Math.min(Number(k.n_gpu_layers)||0,L+1));let m3=(()=>{let V0=k.flash_attn_type&&String(k.flash_attn_type).toLowerCase();if(V0==="on"||V0==="off")return V0;if(V0==="auto")return l.selected.hasGpu?"auto":"off";return l.selected.hasGpu?"auto":"off"})(),c3=$.runtime.cache_dir,B4=H2($,X),B1=await P0(B4,X.size),d3={ok:h,backend:"ggml-llm",warnings:S,errors:C,model:{repoId:X.repoId,revision:X.revision,filename:X.filename,quantization:X.quantization,url:X.url,sizeBytes:X.size,metadata:{architecture:W,n_ctx_train:j,n_layer:L,n_embd:F,quantization_version:q,file_type:_,kv_layer_count:r,swa:b?.enabled?{window:b.window,pattern:b.pattern,dense_first:b.denseFirst,type:b.type,layers:b.swaLayers}:null}},runtime:{...k,variant:l.selected.variant,n_ctx:v,requested_ctx:d,n_gpu_layers:z4,auto_gpu_layers:y0,flash_attn_type:m3,cache_type_k:p.k,cache_type_v:p.v,estimated_max_n_ctx:C0},resources:{modelBytes:f,kvCacheBytes:A1,recurrentMemoryBytes:W0,totalEstimatedBytes:p3,gpuCapacityBytes:L4,gpuUsableBytes:F0,cpuUsableBytes:h0,fit:l.selected.fit},devices:{selected:l.selected,attempts:l.attempts},download:{cacheDir:c3,localPath:B4,exists:B1},timestamp:new Date().toISOString()};return{config:$,info:d3,artifact:X,metadata:{arch:W,nCtxTrain:j,nLayer:L,nEmbd:F},devices:l,cacheTypes:p,localPath:B4,localExists:B1}},P6=($,X,Z=null,W=null)=>{let j,V=Date.now(),Y=0;return new Z2({async start(J){try{let N=await $.parallel.completion(X,(G,w)=>{if(!w)return;if(w.token)Y+=1;J.enqueue({event:"token",data:{requestId:G,...w}})}),{requestId:H}=N;j=N.stop;let O=await N.promise;console.log("[Completion] Result:",O),J.enqueue({event:"result",data:{requestId:H,...O}}),J.close();let q=Date.now()-V,_=O.timings||{};$0.addCompletion({id:`completion-${H}`,generatorId:Z,requestId:H,repoId:W?.repoId||null,quantization:W?.quantization||null,variant:W?.variant||null,cacheTokens:_.cache_n??0,promptTokens:_.prompt_n??0,tokensGenerated:_.predicted_n??Y,tokensPerSecond:_.predicted_per_second??0,promptPerSecond:_.prompt_per_second??0,durationMs:q,success:!0,interrupted:O.interrupted||!1,contextFull:O.context_full||O.contextFull||!1})}catch(N){J.enqueue({event:"error",data:{message:N?.message||String(N)}}),J.error(N),$0.addCompletion({id:`completion-${Date.now()}`,generatorId:Z,repoId:W?.repoId||null,quantization:W?.quantization||null,variant:W?.variant||null,durationMs:Date.now()-V,tokensGenerated:Y,success:!1,error:N?.message||String(N)})}},cancel(){if(j)j()}})},T6=($,X,Z,W,j,V,Y=null,J=null,N=null)=>{let H,O="",q=!1,_=Date.now(),G=0,w=()=>{if(j)o(j).catch(()=>{});if(N)o(N).catch(()=>{})};return new Z2({async start(z){try{let A=await $.parallel.completion(X,(B,L)=>{if(!L)return;if(L.token)O+=L.token,G+=1;z.enqueue({event:"token",data:{requestId:B,...L}})}),{requestId:R}=A;H=A.stop;let U=await A.promise;if(U.text)O=U.text;else if(U.content)O=U.content;q=!U.interrupted&&!U.context_full,console.log("[Completion] Result:",U),z.enqueue({event:"result",data:{requestId:R,...U}}),z.close();let Q=Date.now()-_,K=U.timings||{};if($0.addCompletion({id:`completion-${R}`,generatorId:Y,requestId:R,repoId:J?.repoId||null,quantization:J?.quantization||null,variant:J?.variant||null,cacheTokens:K.cache_n??0,promptTokens:K.prompt_n??V??0,tokensGenerated:K.predicted_n??G,tokensPerSecond:K.predicted_per_second??0,promptPerSecond:K.prompt_per_second??0,durationMs:Q,success:!0,interrupted:U.interrupted||!1,contextFull:U.context_full||U.contextFull||!1,usedCache:Boolean(X.load_state_path)}),q&&Z.enabled&&O)Z.saveCompletionState(W,O,j,V,N).catch((B)=>{console.warn("[SessionCache] Save failed:",B.message)});else w()}catch(A){z.enqueue({event:"error",data:{message:A?.message||String(A)}}),z.error(A),$0.addCompletion({id:`completion-${Date.now()}`,generatorId:Y,repoId:J?.repoId||null,quantization:J?.quantization||null,variant:J?.variant||null,durationMs:Date.now()-_,tokensGenerated:G,success:!1,error:A?.message||String(A)}),w()}},cancel(){if(H)H();w()}})},z0=($)=>{let X={model:$.plan.localPath,runtime:$.plan.info.runtime};return p4("sha256").update(JSON.stringify(X)).digest("hex").slice(0,24)},k6=async($,X,Z,W=null)=>{let{config:j,localPath:V,artifact:Y}=$;if($.localExists&&!X.has(V)){if($.info.download.exists=!0,typeof Z==="function")Z(0.5);return V}if(j.model.local_path&&!j.model.allow_local_file)throw Error("Local model path provided but `model.allow_local_file` is not enabled");let J=V;if(W){let N=W.getDownload(J);if(N){console.log(`[ensureModelFile] Waiting for global download: ${Y.repoId}`);try{if(await N,await P0(V,Y.size)){if($.localExists=!0,$.info.download.exists=!0,typeof Z==="function")Z(0.5);return V}}catch(H){console.warn(`[ensureModelFile] Global download failed, will retry: ${H.message}`)}}}if(!X.has(J))X.set(J,(async()=>{if(Y.isSplit&&Y.splitCount>0){let N=/-(\d{5})-of-(\d{5})\.gguf$/,H=y.dirname(V),O=Y.splitCount,q=0;for(let _=1;_<=O;_+=1){let G=String(_).padStart(5,"0"),w=Y.filename.replace(N,`-${G}-of-${String(O).padStart(5,"0")}.gguf`),z=`${j.model.base_url.replace(/\/+$/,"")}/${Y.repoId}/resolve/${Y.revision}/${w}`,A=y.join(H,w);if(!await P0(A))await a0(z,Y.headers,A,null,(U)=>{if(U>=0&&Number.isFinite(U)){let Q=(q+U)/O,K=Math.round(Q*100);if(console.log(`Downloading model splits: ${Math.min(100,K)}%`),typeof Z==="function")Z(Q*0.5)}});q+=1}}else console.log("Downloading model: 0%"),await a0(Y.url,Y.headers,V,Y.size,(N)=>{if(N>=0&&Number.isFinite(N)){let H=Math.round(N*100);if(console.log(`Downloading model: ${Math.min(100,H)}%`),typeof Z==="function")Z(N*0.5)}});$.localExists=!0,$.info.download.exists=!0})());try{await X.get(J)}finally{X.delete(J)}return V},D6=async($,X)=>{let Z=z0($),W=$.contexts.get(Z);if(W&&!W.released){if(W.releaseTimer)clearTimeout(W.releaseTimer),W.releaseTimer=null,console.log(`[Context] Cancelled pending release for context "${Z}"`);if(W.releaseRequested=!1,W.refCount+=1,console.log(`[Context] Reusing existing context "${Z}", refCount=${W.refCount}`),typeof X==="function")X(0);if(!W.context)await W.ready;if(typeof X==="function")X(1);return W}if(W)console.log(`[Context] Record exists but released=${W.released}, creating new context`);else console.log(`[Context] No existing record for "${Z}", creating new context`);W={key:Z,refCount:1,ready:null,released:!1},$.contexts.set(Z,W),W.ready=(async()=>{let j=Date.now(),V=await k6($.plan,$.downloads,X,$.globalDownloadManager);if(typeof X==="function")X(0.5);let Y={model:V,n_threads:$.plan.info.runtime.n_threads,use_mmap:$.plan.info.runtime.use_mmap,use_mlock:$.plan.info.runtime.use_mlock,cpu_mask:$.plan.info.runtime.cpu_mask,cpu_strict:$.plan.info.runtime.cpu_strict,devices:$.plan.info.runtime.devices,n_ctx:$.plan.info.runtime.n_ctx,n_gpu_layers:$.plan.info.runtime.n_gpu_layers,n_parallel:$.plan.info.runtime.n_parallel,n_batch:$.plan.info.runtime.n_batch,n_ubatch:$.plan.info.runtime.n_ubatch,n_cpu_moe:$.plan.info.runtime.n_cpu_moe,flash_attn_type:$.plan.info.runtime.flash_attn_type,ctx_shift:$.plan.info.runtime.ctx_shift,kv_unified:$.plan.info.runtime.kv_unified,swa_full:$.plan.info.runtime.swa_full,lib_variant:$.plan.info.runtime.variant};if($.plan.info.runtime.flash_attn_type!=="off")Y.cache_type_k=$.plan.info.runtime.cache_type_k,Y.cache_type_v=$.plan.info.runtime.cache_type_v;console.log("[Context] Load Options:",Y);let J;try{if(J=await N6(Y,(N)=>{if(typeof X==="function"){if(X(0.5+N*0.25),N%5===0)console.log("[Context] Load Model Progress:",N)}}),$.plan.info.runtime.n_parallel){if(!await J.parallel.enable({n_parallel:$.plan.info.runtime.n_parallel,n_batch:$.plan.info.runtime.n_batch}))throw Error("Failed to enable parallel decoding mode for context")}if(typeof X==="function")X(1);return W.context=J,W.modelInfo=J.getModelInfo(),$0.addModelLoad({id:`${$.id}-${Z}`,generatorId:$.id,contextKey:Z,repoId:$.plan.info.model?.repoId||null,quantization:$.plan.info.model?.quantization||null,variant:$.plan.info.runtime?.variant||null,nCtx:$.plan.info.runtime?.n_ctx||null,nGpuLayers:$.plan.info.runtime?.n_gpu_layers||null,durationMs:Date.now()-j,success:!0}),W}catch(N){if($0.addModelLoad({id:`${$.id}-${Z}`,generatorId:$.id,contextKey:Z,repoId:$.plan.info.model?.repoId||null,quantization:$.plan.info.model?.quantization||null,variant:$.plan.info.runtime?.variant||null,durationMs:Date.now()-j,success:!1,error:N?.message||String(N)}),J)try{J.release()}catch(H){}throw N}})();try{return await W.ready,W}catch(j){throw $.contexts.delete(Z),j}},r0=async($,X,Z=!1)=>{if(X.released)return!1;if(!Z&&X.refCount>0)return!1;X.released=!0,$.contexts.delete(X.key);try{X.context?.parallel?.disable?.()}catch(W){}return await X.context?.release?.(),!0},b6=async($,X,Z=!1)=>{if(X.releaseRequested=!0,X.releaseTimer)clearTimeout(X.releaseTimer),X.releaseTimer=null;if(Z)X.refCount=0;else if(X.refCount=Math.max(0,X.refCount-1),X.refCount>0)return X.releaseRequested=!1,!1;let W=$.config.runtime.context_release_delay_ms;if(typeof W!=="number"||!Number.isFinite(W))return r0($,X);let j=Math.max(0,Math.floor(W));if(Z||j<=0)return r0($,X);return console.log(`[Context] Scheduling release in ${j}ms for context "${X.key}"`),X.releaseTimer=setTimeout(async()=>{if(X.releaseTimer=null,X.refCount>0){console.log(`[Context] Release cancelled, refCount=${X.refCount} for context "${X.key}"`),X.releaseRequested=!1;return}console.log(`[Context] Releasing context "${X.key}" after ${j}ms delay`),await r0($,X)},j),!0},d4=($)=>{let X=e0($);return X.model.repo_id||X.model.repository||X.model.model||null},a1=($)=>{if(!$)return 0;if(typeof $.score==="number"&&Number.isFinite($.score))return Number($.score);return q0($)};var l4=P(()=>{M0();s0();O6=H6(),{ReadableStream:Z2,WritableStream:K6}=O6,n=y.join(T0.homedir(),".buttress","models"),Q2=["mxfp4","q8_0","q6_k","q6","q5_k_m","q5_k_s","q5_k","q5_1","q5_0","q4_k_m","q4_k_s","q4_k","q4_1","q4_0","q3","q2"],S0={backend:{type:"ggml-llm",variant:null,variant_preference:["cuda","vulkan","snapdragon","default"],gpu_memory_fraction:0.85,cpu_memory_fraction:t0},model:{repo_id:null,revision:"main",filename:null,url:null,quantization:null,preferred_quantizations:[],n_ctx:null,n_gpu_layers:"auto",allow_local_file:!1,local_path:null,api_base:Y2,base_url:j2},runtime:{cache_dir:n,prefer_variants:[],huggingface_token:process.env.HUGGINGFACE_TOKEN||null,http_headers:{},session_cache:{enabled:!0,max_size_bytes:10737418240,max_entries:1000},context_release_delay_ms:1e4}}});import j0 from"node:path";import t4 from"node:os";import{stat as B2,mkdir as h6,open as I6,unlink as i4,readFile as C6,writeFile as y6}from"node:fs/promises";import{createHash as u6}from"node:crypto";import{initWhisper as f6}from"@fugood/whisper.node";import{getBackendDevicesInfo as R2,isLibVariantAvailable as w2}from"@fugood/llama.node";import*as g6 from"node:stream/web";class v2{constructor(){this.queue=[],this.processing=!1,this.currentTaskId=null}async enqueue($,X=null){return new Promise((Z,W)=>{this.queue.push({task:$,resolve:Z,reject:W,taskId:X}),this.processNext()})}async processNext(){if(this.processing||this.queue.length===0)return;this.processing=!0;let{task:$,resolve:X,reject:Z,taskId:W}=this.queue.shift();this.currentTaskId=W;try{let j=await $();X(j)}catch(j){Z(j)}finally{this.processing=!1,this.currentTaskId=null,this.processNext()}}getStatus(){return{processing:this.processing,queuedCount:this.queue.length,currentTaskId:this.currentTaskId}}}async function h2($,X,Z={}){let{globalDownloadManager:W=null}=Z,j=j4(X),V=await o6(j),Y={id:$,type:"ggml-stt",config:j,plan:V,info:V.info,contextRecord:null,downloads:new Map,globalDownloadManager:W,queue:new v2,finalized:!1},J=async()=>{if(Y.finalized)return;Y.finalized=!0;let A=Y.contextRecord;if(!A)return;if(A.released)return;if(A.releaseRequested||A.releaseTimer)return;if(A.refCount=Math.max(0,A.refCount-1),A.refCount>0)return;await Z4(Y,A)},N=async(A={})=>{let{onProgress:R}=A;try{let U=await e6(Y,R);return{modelInfo:U.modelInfo&&typeof U.modelInfo==="object"?{...U.modelInfo}:null,runtime:{...Y.plan.info.runtime},download:{...Y.plan.info.download}}}catch(U){throw console.error("[Context] Error initializing context:",U),U}},H=async()=>{if(Y.finalized)return!1;let A=Y.contextRecord;if(!A)return!1;return $9(Y,A)},O=async(A={})=>{let{audioPath:R,audioData:U,options:Q={}}=A,K=Y.contextRecord;if(!K)throw Error("Context not initialized");let B={...Q};if(Y.plan.info.runtime.max_threads&&B.maxThreads==null)B.maxThreads=Y.plan.info.runtime.max_threads;let L=`transcription-${Date.now()}-${Math.random().toString(36).slice(2,8)}`,F=Date.now();return Y.queue.enqueue(async()=>{await K.ready;try{let E;if(U){let M=t6(U),{promise:x}=K.context.transcribeData(M,B);E=await x}else{if(!R)throw Error("audioPath or audioData is required for transcription");let M=j0.resolve(R),{promise:x}=K.context.transcribe(M,B);E=await x}return N0.addTranscription({id:L,generatorId:Y.id,repoId:Y.plan.info.model?.repoId||null,quantization:Y.plan.info.model?.quantization||null,modelType:Y.plan.info.model?.modelType||null,variant:Y.plan.info.runtime?.variant||null,durationMs:Date.now()-F,segmentCount:E?.segments?.length||0,textLength:E?.text?.length||0,success:!0}),E}catch(E){throw N0.addTranscription({id:L,generatorId:Y.id,repoId:Y.plan.info.model?.repoId||null,quantization:Y.plan.info.model?.quantization||null,modelType:Y.plan.info.model?.modelType||null,variant:Y.plan.info.runtime?.variant||null,durationMs:Date.now()-F,success:!1,error:E?.message||String(E)}),E}},L)},q=async(A={})=>O(A),_=async(A={})=>O(A),G=()=>{let A=Y.contextRecord;if(!A)return!1;return!A.released&&(A.releaseRequested||A.releaseTimer||A.refCount>0)},w=()=>{Y.finalized=!1},z=()=>({id:Y.id,type:Y.type,repoId:Y.plan.info.model?.repoId||null,quantization:Y.plan.info.model?.quantization||null,modelType:Y.plan.info.model?.modelType||null,variant:Y.plan.info.runtime?.variant||null,hasContext:Boolean(Y.contextRecord?.context),contextRefCount:Y.contextRecord?.refCount||0,queueStatus:Y.queue.getStatus()});return{id:$,type:"ggml-stt",info:V.info,queue:Y.queue,initContext:N,transcribe:q,transcribeData:_,releaseContext:H,finalize:J,getStatus:z,hasPendingReleases:G,resetFinalized:w}}async function I2($,X,Z={}){let{onProgress:W,onComplete:j,onError:V}=Z;try{let Y=j4($),J=await e4(Y),N=D2(Y,J),{repoId:H}=J;if(await W4(N,J.size)){if(console.log(`[Download] STT model already exists: ${H} at ${N}`),typeof j==="function")j({localPath:N,repoId:H,alreadyExists:!0});return{started:!1,localPath:N,repoId:H,alreadyExists:!0}}let q=X.getDownload(N);if(q)return console.log(`[Download] Already downloading STT model: ${H}`),q.then(()=>{if(typeof j==="function")j({localPath:N,repoId:H,joinedExisting:!0})}).catch((G)=>{if(typeof V==="function")V(G)}),{started:!1,localPath:N,repoId:H,alreadyDownloading:!0};console.log(`[Download] Starting STT model download: ${H}`);let _=(async()=>{try{if(await b2(J.url,J.headers,N,J.size,(G)=>{if(G>=0&&Number.isFinite(G)){if(console.log(`[Download] ${H}: ${Math.round(G*100)}%`),typeof W==="function")W(G)}}),console.log(`[Download] Completed STT model: ${H}`),typeof j==="function")j({localPath:N,repoId:H})}catch(G){if(console.error(`[Download] Failed STT model: ${H}`,G.message),typeof V==="function")V(G);throw G}finally{X.deleteDownload(N)}})();return X.setDownload(N,_),{started:!0,localPath:N,repoId:H}}catch(Y){if(console.error("[Download] Failed to start STT download:",Y.message),typeof V==="function")V(Y);return{started:!1,localPath:null,repoId:null,error:Y.message}}}async function C2($=null,X={}){let{threshold:Z=1.1,includeBreakdown:W=!1,config:j,...V}=X,Y=null,J=null,N=null;if(j)try{let R=j4(j),U=await e4(R);Y=U.size??null,{processingBufferBytes:J}=m0({modelBytes:Y}),N=U.quantization||null}catch(R){}let H=j?.backend?.gpu_memory_fraction!=null?Math.min(1,Math.max(0,Number(j.backend.gpu_memory_fraction))):void 0,O=j?.backend?.cpu_memory_fraction!=null?Math.min(1,Math.max(0,Number(j.backend.cpu_memory_fraction))):void 0,q=await _0({...V,platform:process.platform,totalMemoryInBytes:t4.totalmem(),backend:"ggml-stt",includeBreakdown:W,gpuMemoryFraction:H,cpuMemoryFraction:O,dependencies:{getBackendDevicesInfo:R2,isLibVariantAvailable:w2},modelBytes:Y,kvCacheBytes:J}),_=q.selected,G=z2(_);if(_)_.modelBytes=Y||null,_.processingBytes=J||null,_.quantization=N||null;let w=null,z=null;if($){let R=z2($);z={...$,score:R};let U="buttress",Q="buttress-higher-score";if(!q.ok)U="local",Q="buttress-unavailable";else if(!R&&R!==0)U="buttress",Q="missing-client-score";else if($.fit&&_?.fit){let K=$.fit.fitsInGpu||$.fit.fitsInCpu,B=_.fit.fitsInGpu||_.fit.fitsInCpu;if(K&&!B)U="local",Q="client-fits-in-memory";else if(B&&!K)U="buttress",Q="buttress-fits-in-memory";else if(R>G*Z)U="local",Q="client-better";else if(G>R*Z)U="buttress",Q="buttress-better";else U="either",Q="comparable-scores"}else if(R>G*Z)U="local",Q="client-better";else if(G>R*Z)U="buttress",Q="buttress-better";else U="either",Q="comparable-scores";w={buttressScore:G,clientScore:R,threshold:Z,recommendation:U,reason:Q}}if(!q.ok&&!w)w={buttressScore:G,clientScore:$?.score??null,threshold:Z,recommendation:"local",reason:"buttress-unavailable"};let A=null;if(j)A={repoId:j.model?.repo_id||null,quantization:j.model?.quantization||null,filename:j.model?.filename||null};return{type:"ggml-stt",timestamp:new Date().toISOString(),buttress:q,client:z,comparison:w,modelConfig:A}}var p6=()=>{if(typeof globalThis<"u"&&globalThis.ReadableStream&&globalThis.WritableStream)return{ReadableStream:globalThis.ReadableStream,WritableStream:globalThis.WritableStream};return g6},m6,F2=($={},X={})=>{return Object.entries(X||{}).forEach(([Z,W])=>{if(W&&typeof W==="object"&&!Array.isArray(W)){if(!$[Z]||typeof $[Z]!=="object")$[Z]={};F2($[Z],W)}else $[Z]=W}),$},c6=".bin",E2="https://huggingface.co",M2="https://huggingface.co/api",R0,r4,o4,x2="fp16",S2=0.5,d6,P2=($)=>{if(!$)return null;let X=$.toLowerCase();return d6.find((Z)=>X.includes(Z))||null},a4,s4=($,X=[])=>{if(!$&&$!==0)return[...X];if(Array.isArray($))return $.filter((Z)=>Z!=null);return[$]},X4=($)=>{if(!$)return null;let X=String($).toLowerCase();if(["cuda","vulkan","default"].includes(X))return X;return null},j4=($={})=>{let X=JSON.parse(JSON.stringify(a4));if(F2(X,$),X.backend.variant=X4(X.backend.variant),X.backend.variant_preference=Array.from(new Set(s4(X.backend.variant_preference||r4).map(X4).filter(Boolean))),X.backend.variant_preference.length===0)X.backend.variant_preference=[...r4];if(X.runtime.prefer_variants=Array.from(new Set(s4(X.runtime.prefer_variants).map(X4).filter(Boolean))),X.model.preferred_quantizations=Array.from(new Set(s4(X.model.preferred_quantizations||X.model.quantizations).map((Z)=>Z?String(Z).toLowerCase():null).filter(Boolean))),X.model.quantization){let Z=String(X.model.quantization).toLowerCase();if(!X.model.preferred_quantizations.includes(Z))X.model.preferred_quantizations.unshift(Z)}return X.model.base_url=X.model.base_url||E2,X.model.api_base=X.model.api_base||M2,X.runtime.cache_dir=X.runtime.cache_dir?j0.resolve(X.runtime.cache_dir):R0,X.runtime.context_release_delay_ms=Math.max(0,Number(X.runtime.context_release_delay_ms)||a4.runtime.context_release_delay_ms),X},n4=($)=>{let X=$.toLowerCase();return o4.find((W)=>X.includes(W))||null},l6=($)=>{let X=[];if($.backend.variant)X.push($.backend.variant);if($.runtime.prefer_variants.length>0)X.push(...$.runtime.prefer_variants);return X.push(...$.backend.variant_preference),X.push("default"),Array.from(new Set(X.map(X4).filter(Boolean)))},T2=async($)=>{await h6($,{recursive:!0})},i6=($=R0)=>j0.join($,".metadata-cache"),k2=($,X,Z=R0)=>{let W=u6("sha256").update($).digest("hex");return j0.join(i6(Z),X,`${W}.json`)},s6=async($,X,Z=R0)=>{try{let W=k2($,X,Z),j=await C6(W,"utf-8");return JSON.parse(j)}catch(W){return null}},L2=async($,X,Z,W=R0)=>{try{let j=k2($,X,W);await T2(j0.dirname(j)),await y6(j,JSON.stringify(Z),"utf-8")}catch(j){}},n6=async($,X={})=>{if(typeof fetch!=="function")throw Error("Global fetch is not available in this runtime");let Z=await fetch($,X);if(!Z.ok){let W=await Z.text().catch(()=>"");throw Error(`Failed to fetch ${$}: ${Z.status} ${Z.statusText} ${W}`.trim())}return Z.json()},A2=async($,X={})=>{if(typeof fetch!=="function")throw Error("Global fetch is not available in this runtime");let Z=await fetch($,{...X,method:"HEAD"});if(!Z.ok)throw Error(`Failed to fetch headers for ${$}: ${Z.status} ${Z.statusText}`);return Z},D2=($,X)=>{if($.model.local_path)return j0.resolve($.model.local_path);let Z=X.repoId.split("/"),W=j0.join($.runtime.cache_dir,...Z,X.revision);return j0.join(W,X.filename)},W4=async($,X)=>{try{let Z=await B2($);if(!X)return!0;return Z.size===X}catch(Z){return!1}},b2=async($,X,Z,W,j)=>{if(typeof fetch!=="function")throw Error("Global fetch is not available in this runtime");await T2(j0.dirname(Z));let V=await fetch($,{headers:X});if(!V.ok||!V.body)throw Error(`Failed to download ${$}: ${V.status} ${V.statusText}`);let Y=await I6(Z,"w"),J=Number(V.headers.get("content-length"))||W||0,N=0,H=0.05;try{await V.body.pipeTo(new m6({async write(O){if(await Y.write(O),N+=O.byteLength,typeof j==="function"&&J>0){let q=Math.min(1,N/J);while(q>=H)j(H),H+=0.05}},async close(){if(await Y.close(),typeof j==="function")j(1)},async abort(O){throw await Y.close().catch(()=>{}),await i4(Z).catch(()=>{}),O}}))}catch(O){throw await Y.close().catch(()=>{}),await i4(Z).catch(()=>{}),O}if(W){let O=await B2(Z);if(O.size!==W)throw await i4(Z).catch(()=>{}),Error(`Downloaded file size mismatch, expected ${W} got ${O.size}`)}},e4=async($)=>{let X=$.model.repo_id||$.model.repository||$.model.model;if(!X)throw Error("`model.repo_id` is required in Buttress backend config");let Z=$.model.revision||"main",W=$.runtime.cache_dir,j=JSON.stringify({repoId:X,revision:Z,filename:$.model.filename,url:$.model.url,quantization:$.model.quantization,preferred_quantizations:$.model.preferred_quantizations}),V=await s6(j,"artifact-info",W);if(V)return V;let Y={...$.runtime.http_headers||{}};if($.runtime.huggingface_token)Y.Authorization=`Bearer ${$.runtime.huggingface_token}`;if($.model.url){let U=await A2($.model.url,{headers:Y}),Q=Number(U.headers.get("content-length"))||null,K=$.model.filename||$.model.url.split("/").pop(),B={repoId:X,revision:Z,filename:K,url:$.model.url,size:Q,quantization:n4(K||""),headers:Y};return await L2(j,"artifact-info",B,W),B}let{filename:J}=$.model,N=$.model.quantization&&String($.model.quantization).toLowerCase(),H=await n6(`${$.model.api_base}/models/${X}?revision=${Z}&blobs=true`,{headers:Y}),q=(H?.siblings||H?.files||[]).map((U)=>U.rfilename||U.path||U.filename).filter((U)=>typeof U==="string"&&U.endsWith(c6));if(q.length===0)throw Error(`No model artifacts found in repo ${X}`);let _=$.model.preferred_quantizations.length>0?$.model.preferred_quantizations:o4,G=()=>{for(let U of _)if(U===x2){let Q=q.find((K)=>{let B=K.toLowerCase();return!o4.some((L)=>B.includes(L))});if(Q)return{filename:Q,quantization:null}}else{let Q=q.find((K)=>K.toLowerCase().includes(U));if(Q)return{filename:Q,quantization:U}}return null};if(!J){let U=G()||{filename:q[0],quantization:null},{filename:Q,quantization:K}=U;J=Q,N=K||n4(J)}else if(!N)N=n4(J);let w=`${$.model.base_url.replace(/\/+$/,"")}/${X}/resolve/${Z}/${J}`,z=await A2(w,{headers:Y}),A=Number(z.headers.get("content-length"))||null,R={repoId:X,revision:Z,filename:J,url:w,size:A,quantization:N,headers:Y,isSplit:!1,splitCount:0};return await L2(j,"artifact-info",R,W),R},r6=async($,{modelBytes:X=null,processingBytes:Z=null}={})=>{let W=l6($),[j,...V]=W,Y=$.backend?.gpu_memory_fraction!=null?Math.min(1,Math.max(0,Number($.backend.gpu_memory_fraction))):a4.backend.gpu_memory_fraction||1,J=$.backend?.cpu_memory_fraction!=null?Math.min(1,Math.max(0,Number($.backend.cpu_memory_fraction))):S2,N=await _0({platform:process.platform,totalMemoryInBytes:t4.totalmem(),backend:"ggml-stt",variant:j||null,preferVariants:V,variantPreference:$.backend.variant_preference,gpuMemoryFraction:Y,cpuMemoryFraction:J,dependencies:{getBackendDevicesInfo:R2,isLibVariantAvailable:w2},modelBytes:X,kvCacheBytes:Z}),H=(_)=>({..._,devices:Array.isArray(_.devices)?_.devices:[],ok:_.ok,hasGpu:Boolean(_.hasGpu),totalMemory:_.gpuTotalBytes||_.totalMemory||0,error:_.ok?null:Error(_.error||`Variant ${_.variant} not available on this platform`)});if(!N.ok||!N.selected){let _=(N.attempts||[]).map((G)=>`${G.variant}: ${G.error||"unknown error"}`).join("; ");throw Error(`Unable to initialize any backend variant (${W.join(", ")}). Errors: ${_}`)}let O=(N.attempts||[]).map(H);return{selected:H(N.selected),attempts:O}},o6=async($)=>{let X=await e4($),Z=m0({modelBytes:X.size>0?X.size:0}),W=await r6($,{modelBytes:Z.modelBytes,processingBytes:Z.processingBufferBytes}),j=W.selected.hasGpu&&(W.selected.fit?.fitsInGpu!==void 0?W.selected.fit.fitsInGpu:!0);if($.model.use_gpu===!1)j=!1;let V=$.model.use_flash_attn&&String($.model.use_flash_attn).toLowerCase(),Y;if(V==="on"||V==="true")Y=!0;else if(V==="off"||V==="false")Y=!1;else Y=j;let J=$.runtime.cache_dir,N=D2($,X),H=await W4(N,X.size),O={ok:!0,backend:"ggml-stt",model:{repoId:X.repoId,revision:X.revision,filename:X.filename,quantization:X.quantization,modelType:P2(X.filename),url:X.url,sizeBytes:X.size},runtime:{variant:W.selected.variant,use_gpu:j,use_flash_attn:Y,max_threads:$.runtime.max_threads?Number($.runtime.max_threads):null},resources:{...Z,gpuCapacityBytes:W.selected.gpuTotalBytes,gpuUsableBytes:W.selected.gpuUsableBytes,cpuUsableBytes:W.selected.cpuUsableBytes,fit:W.selected.fit},devices:{selected:W.selected,attempts:W.attempts},download:{cacheDir:J,localPath:N,exists:H},timestamp:new Date().toISOString()};return{config:$,info:O,artifact:X,memory:Z,devices:W,localPath:N,localExists:H}},a6=async($,X,Z,W=null)=>{let{localPath:j,artifact:V,config:Y}=$;if($.localExists){if(typeof Z==="function")Z(1);return j}if(W){let H=W.getDownload(j);if(H){console.log(`[ensureModelFile] Waiting for global STT download: ${V.repoId}`);try{if(await H,await W4(j,V.size)){if($.localExists=!0,$.info.download.exists=!0,typeof Z==="function")Z(1);return j}}catch(O){console.warn(`[ensureModelFile] Global STT download failed, will retry: ${O.message}`)}}}let J=X.get(j);if(J){if(await J,typeof Z==="function")Z(1);return j}let N=(async()=>{if(Y.model.allow_local_file){if(!await W4(j,V.size))throw Error(`Local model file not found: ${j}`);return j}return await b2(V.url,V.headers,j,V.size,Z),j})();X.set(j,N);try{return await N,j}finally{X.delete(j)}},t6=($)=>{if(!$)return null;if($ instanceof ArrayBuffer)return $;if(ArrayBuffer.isView($))return $.buffer;if(typeof $==="string"){let X=$.startsWith("data:")?$.split(",")[1]||"":$,Z=Buffer.from(X,"base64");return Z.buffer.slice(Z.byteOffset,Z.byteOffset+Z.byteLength)}throw Error("Unsupported audioData format, expected base64 string or ArrayBuffer")},e6=async($,X)=>{if($.contextRecord&&!$.contextRecord.released){if($.contextRecord.releaseTimer)clearTimeout($.contextRecord.releaseTimer),$.contextRecord.releaseTimer=null,console.log("[Context] Cancelled pending STT release");if($.contextRecord.releaseRequested=!1,$.contextRecord.refCount+=1,console.log(`[Context] Reusing existing STT context, refCount=${$.contextRecord.refCount}`),typeof X==="function")X(0);if(!$.contextRecord.context)await $.contextRecord.ready;if(typeof X==="function")X(1);return $.contextRecord}if($.contextRecord)console.log(`[Context] STT record exists but released=${$.contextRecord.released}, creating new context`);else console.log("[Context] No existing STT record, creating new context");let Z={refCount:1,ready:null,released:!1};$.contextRecord=Z,Z.ready=(async()=>{let W=Date.now();try{if(typeof X==="function")X(0);let j=await a6($.plan,$.downloads,X,$.globalDownloadManager);if(typeof X==="function")X(0.5);let V=await f6({filePath:j,useFlashAttn:$.plan.info.runtime.flash_attn_type==="on",useGpu:$.plan.info.runtime.n_gpu_layers>0,nThreads:$.plan.info.runtime.n_threads},$.plan.info.runtime.variant);if(typeof X==="function")X(1);Z.context=V;try{Z.modelInfo=V.getModelInfo()}catch(Y){Z.modelInfo=null}return N0.addModelLoad({id:$.id,generatorId:$.id,repoId:$.plan.info.model?.repoId||null,quantization:$.plan.info.model?.quantization||null,modelType:$.plan.info.model?.modelType||null,variant:$.plan.info.runtime?.variant||null,useGpu:$.plan.info.runtime?.use_gpu||!1,durationMs:Date.now()-W,success:!0}),Z}catch(j){throw N0.addModelLoad({id:$.id,generatorId:$.id,repoId:$.plan.info.model?.repoId||null,quantization:$.plan.info.model?.quantization||null,modelType:$.plan.info.model?.modelType||null,variant:$.plan.info.runtime?.variant||null,durationMs:Date.now()-W,success:!1,error:j?.message||String(j)}),j}})();try{if(await Z.ready,typeof X==="function")X(1);return Z}catch(W){throw $.contextRecord=null,W}},Z4=async($,X,Z=!1)=>{if(X.released)return!1;if(!Z&&X.refCount>0)return!1;return X.released=!0,$.contextRecord=null,await X.context?.release?.(),!0},$9=async($,X,Z=!1)=>{if(X.releaseRequested=!0,X.releaseTimer)clearTimeout(X.releaseTimer),X.releaseTimer=null;if(Z)X.refCount=0;else if(X.refCount=Math.max(0,X.refCount-1),X.refCount>0)return X.releaseRequested=!1,!1;let W=$.config.runtime.context_release_delay_ms;if(typeof W!=="number"||!Number.isFinite(W))return Z4($,X);let j=Math.max(0,Math.floor(W));if(Z||j<=0)return Z4($,X);return console.log(`[Context] Scheduling STT release in ${j}ms`),X.releaseTimer=setTimeout(async()=>{if(X.releaseTimer=null,X.refCount>0){console.log(`[Context] STT release cancelled, refCount=${X.refCount}`),X.releaseRequested=!1;return}console.log(`[Context] Releasing STT context after ${j}ms delay`),await Z4($,X)},j),!0},$1=($)=>{let X=j4($),Z=X.model.repo_id||X.model.repository||X.model.model||null;if(!Z)return null;let W=P2(X.model.filename);if(W)return`${Z}:${W}`;return Z},z2=($)=>{if(!$)return 0;if(typeof $.score==="number"&&Number.isFinite($.score))return Number($.score);return q0($)};var X1=P(()=>{M0();s0();({WritableStream:m6}=p6()),R0=j0.join(t4.homedir(),".buttress","models"),r4=["cuda","vulkan","default"],o4=["q8_0","q5_1","q5_0","q4_1","q4_0"],d6=["large-v3-turbo","distil-large-v3","large-v3","large-v2","large-v1","large","distil-medium","medium.en","medium","small.en-tdrz","distil-small.en","small.en","small","base.en","base","tiny.en","tiny"],a4={backend:{type:"ggml-stt",variant:null,variant_preference:r4,gpu_memory_fraction:0.85,cpu_memory_fraction:S2},model:{repo_id:"BricksDisplay/whisper-ggml",revision:"main",filename:null,url:null,quantization:null,preferred_quantizations:["q8_0",x2,"q5_1"],allow_local_file:!1,local_path:null,api_base:M2,base_url:E2,use_gpu:!0,use_flash_attn:"auto"},runtime:{cache_dir:R0,prefer_variants:[],huggingface_token:process.env.HUGGINGFACE_TOKEN||null,http_headers:{},max_threads:null,context_release_delay_ms:1e4}}});async function t($,X=null,Z={}){if($==="ggml-llm")return _2(X,Z);if($==="ggml-stt")return C2(X,Z);throw Error(`Unknown backend type: ${$}`)}var Z1=P(()=>{l4();X1()});var g;var y2=P(()=>{g={name:"@fugood/buttress-backend-core",private:!0,type:"module",version:"2.23.0-beta.53",main:"src/index.js",types:"lib/types/index.d.ts",scripts:{build:"tsc --noResolve --noCheck --declaration --emitDeclarationOnly --allowJs --outDir lib/types src/index.js"},dependencies:{"@fugood/buttress-hardware-guardrails":"^2.23.0-beta.53","@fugood/llama.node":"^1.6.0-rc.0","@fugood/whisper.node":"^1.0.13","@huggingface/gguf":"^0.3.2","@iarna/toml":"^3.0.0",bytes:"^3.1.0"}}});import m from"node:os";import u2 from"node:fs";import f2 from"node:path";import{execSync as Y4}from"node:child_process";import g2 from"@iarna/toml";async function m2({modelIds:$=[],defaultConfig:X=null}={}){let Z=[];if(console.log(`${g.name} v${g.version}`),console.log(`Generating model capabilities comparison...
+`),Z.push(`${g.name} v${g.version}`),Z.push(`## Model Capabilities Comparison
+`),!$||$.length===0)console.error("Error: No model IDs provided"),process.exit(1);try{let W=(U={},Q={})=>{let K=Array.isArray(U)?[...U]:{...U};return Object.entries(Q||{}).forEach(([B,L])=>{if(L&&typeof L==="object"&&!Array.isArray(L))K[B]=W(K[B]||{},L);else K[B]=L}),K},j=X||{},{server:V,generators:Y=[],...J}=j,N=(U)=>W(JSON.parse(JSON.stringify(J)),U||{}),H=(U)=>{if(Array.isArray(Y)&&Y.length>0){let Q=Y.filter((K)=>K?.type==="ggml-llm");if(Q.length>0&&U){let K=Q.find((B)=>B.model?.repo_id===U);if(K)return N(K)}}return Object.keys(J).length>0?N({}):null},O=[];for(let U=0;U<$.length;U+=1){let Q=$[U];console.log(`[${U+1}/${$.length}] Analyzing ${Q}...`);let K=H(Q);K={...K||{},model:{...J.runtime,...K?.model||{},repo_id:Q}};let B=await t("ggml-llm",null,{config:K,includeBreakdown:!0});O.push({modelId:Q,capabilities:B,modelInfo:B.buttress?.selected||null,modelConfig:B.modelConfig||null})}let q=(U)=>U?(U/1024/1024/1024).toFixed(2):"N/A",_=(U)=>U?"✅":"\uD83D\uDEAB";Z.push("| Model ID | Quantization | Size (GB) | Context Size | KV Cache Size (GB) | Total Required Memory (GB) | Fits GPU (Full) | Fits CPU (Full) |"),Z.push("|----------|--------------|-----------|--------------|--------------------|-----------------------------|-----------------|-----------------|"),O.forEach(({modelId:U,modelInfo:Q,modelConfig:K})=>{let B=Q?.quantization?.name?.toUpperCase()||"N/A",L=q(Q?.modelBytes),F=K?.nCtx||Q?.kvInfo?.nCtxTrain||"N/A",E=G0(Q),M=Number(F),x=Q?.kvCacheBytes||(E&&Number.isFinite(M)&&M>0?E(M):E&&E(Q?.kvInfo?.nCtxTrain||0))||null,D=q(x),I=q(Q?.modelBytes&&x?Q.modelBytes+x:Q?.fit?.totalRequiredBytes),b=_(Q?.fit?.fitsInGpu),u=_(Q?.fit?.fitsInCpu);Z.push(`| ${U} | ${B} | ${L} | ${F} | ${D} | ${I} | ${b} | ${u} |`);let r=Q?.memoryLimitedCtx!=null||Q?.limitedFit!=null,k=!Q?.fit?.fitsInGpu||!Q?.fit?.fitsInCpu;if(r&&k){let d=Q?.memoryLimitedCtx||F,v=Number(d),S=Q?.limitedKvCacheBytes||E&&Number.isFinite(v)&&v>0&&E(v)||null,C=q(S),h=q(Q?.modelBytes&&S?Q.modelBytes+S:Q?.limitedFit?.totalRequiredBytes),p=_(Q?.limitedFit?.fitsInGpu),f=_(Q?.limitedFit?.fitsInCpu);if(d!==F||C!==D||h!==I)Z.push(`| ↳ Limited | - | ${L} | ${d} | ${C} | ${h} | ${p} | ${f} |`)}}),Z.push(`
+---`),Z.push(`
+### System Information`);let G=null;if(process.platform!=="win32")try{G=Y4("uname -a",{encoding:"utf8"}).trim()}catch{}if(G)Z.push(`- **System:** ${G}`);else Z.push(`- **Hostname:** ${m.hostname()}`),Z.push(`- **OS:** ${m.type()} ${m.release()}`);if(Z.push(`- **Platform:** ${process.platform}`),Z.push(`- **CPU Cores:** ${m.cpus().length}`),Z.push(`- **Total System Memory:** ${(m.totalmem()/1024/1024/1024).toFixed(2)} GB`),O.length>0){let Q=O[0].capabilities.buttress?.selected;if(Q){let K=Q.cpuTotalBytes>0?(Q.cpuUsableBytes/Q.cpuTotalBytes*100).toFixed(0):0;if(Z.push(`- **Usable CPU Memory:** ${(Q.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${K}% of ${(Q.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),Q.hasGpu){let B=Q.devices.filter((L)=>L.type==="gpu");if(B.length>0){let L=B[0];Z.push(`- **GPU Backend:** ${L.backend}`),Z.push(`- **GPU Name:** ${L.deviceName}`),Z.push(`- **GPU Total Memory:** ${(L.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let F=Q.gpuTotalBytes>0?(Q.gpuUsableBytes/Q.gpuTotalBytes*100).toFixed(0):0;Z.push(`- **GPU Usable Memory:** ${(Q.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${F}% of ${(Q.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`)}}else Z.push("- **GPU:** Not available")}}Z.push(`
+### Command Used`);let w=process.argv.slice(2).join(" ");if(Z.push(`\`\`\`bash
+${process.argv[0]} ${process.argv[1]} ${w}
+\`\`\``),Z.push(`
+### Package Information`),Z.push(`- **Name:** ${g.name}`),Z.push(`- **Version:** ${g.version}`),g.description)Z.push(`- **Description:** ${g.description}`);if(X&&Object.keys(X).length>0){Z.push(`
+### Configuration`),Z.push("<details>"),Z.push("<summary>Click to expand TOML configuration</summary>"),Z.push("\n```toml");try{let U=g2.stringify(X);Z.push(U)}catch(U){Z.push("# Error serializing config"),Z.push(JSON.stringify(X,null,2))}Z.push("```"),Z.push("</details>")}let A=`ggml-llm-model-capabilities-${new Date().toISOString().replace(/[.:]/g,"-").split("T")[0]}.md`,R=f2.join(process.cwd(),A);u2.writeFileSync(R,Z.join(`
+`),"utf8"),console.log(`
+Model capabilities table saved to: ${R}`),process.exit(0)}catch(W){console.error("Failed to generate model table:",W.message),process.exit(1)}}async function c2({modelId:$=null,defaultConfig:X=null}={}){if(console.log(`${g.name} v${g.version}`),console.log("Testing capabilities for backend: ggml-llm"),$)console.log(`Model: ${$}`);console.log("--------------------------------");try{let Z=X||{},{server:W,generators:j=[],...V}=Z,Y=(G={},w={})=>{let z=Array.isArray(G)?[...G]:{...G};return Object.entries(w||{}).forEach(([A,R])=>{if(R&&typeof R==="object"&&!Array.isArray(R))z[A]=Y(z[A]||{},R);else z[A]=R}),z},J=(G)=>Y(JSON.parse(JSON.stringify(V)),G||{}),H=((G)=>{if(Array.isArray(j)&&j.length>0){let w=j.filter((z)=>z?.type==="ggml-llm");if(w.length>0){if(G){let z=w.find((A)=>A.model?.repo_id===G);if(z)return J(z)}}}if(Object.keys(V).length>0)return J({});return null})($);if($)H={...H||{},model:{...H?.model||{},repo_id:$}};let O=await t("ggml-llm",null,{config:H,includeBreakdown:!0}),q=O.buttress?.selected||null,_=O.modelConfig||null;if($||_?.repoId){console.log(`
+=== Model Information ===`);let G=$||_?.repoId;if(console.log(`Repository ID: ${G}`),_?.quantization)console.log(`Quantization: ${_.quantization}`);if(_?.nCtx)console.log(`Context Length: ${_.nCtx}`);if(q?.quantization?.name)console.log(`Model Quantization: ${q.quantization.name.toUpperCase()}`);let w=_?.cache_type_k||"f16",z=_?.cache_type_v||"f16";if(console.log(`KV Cache Type: K=${w}, V=${z}`),q?.modelBytes&&q?.kvCacheBytes){if(console.log(`Model Size: ${(q.modelBytes/1024/1024/1024).toFixed(2)} GB`),q.kvInfo)console.log(`KV Cache Size: ${(q.kvCacheBytes/1024/1024/1024).toFixed(2)} GB (KV info: ${JSON.stringify(q.kvInfo)})`);else console.log(`KV Cache Size: ${(q.kvCacheBytes/1024/1024/1024).toFixed(2)} GB`);if(console.log(`Total Required Memory: ${((q.modelBytes+q.kvCacheBytes)/1024/1024/1024).toFixed(2)} GB`),q.memoryLimitedCtx!=null){let A=q.memoryLimitedCtx,R=q.kvInfo?.nCtxTrain;if(R)console.log(`
+Memory-Limited Context: ${A} (Train: ${R})`);else console.log(`
+Memory-Limited Context: ${A}`);if(q.limitedKvCacheBytes!=null)console.log(`Limited KV Cache Size: ${(q.limitedKvCacheBytes/1024/1024/1024).toFixed(2)} GB`)}}else if(O.buttress?.selected?.fit){let{totalRequiredBytes:A}=O.buttress.selected.fit;console.log(`Total Required Memory: ${(A/1024/1024/1024).toFixed(2)} GB`)}}if(O.buttress?.selected){let{selected:G}=O.buttress;console.log(`
+=== Hardware Information ===`);let w=null;if(process.platform!=="win32")try{w=Y4("uname -a",{encoding:"utf8"}).trim()}catch{}if(w)console.log(`System: ${w}`);else console.log(`Hostname: ${m.hostname()}`),console.log(`OS: ${m.type()} ${m.release()}`);console.log(`Platform: ${G.platform}`),console.log(`CPU Cores: ${m.cpus().length}`),console.log(`Total System Memory: ${(m.totalmem()/1024/1024/1024).toFixed(2)} GB`);let z=G.cpuTotalBytes>0?(G.cpuUsableBytes/G.cpuTotalBytes*100).toFixed(0):0;if(console.log(`Usable CPU Memory: ${(G.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${z}% of ${(G.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),G.hasGpu)console.log(`
+--- GPU Details ---`),G.devices.filter((R)=>R.type==="gpu").forEach((R)=>{console.log(`GPU Backend: ${R.backend}`),console.log(`GPU Name: ${R.deviceName}`),console.log(`GPU Total Memory: ${(R.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let U=G.gpuTotalBytes>0?(G.gpuUsableBytes/G.gpuTotalBytes*100).toFixed(0):0;if(console.log(`GPU Usable Memory: ${(G.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${U}% of ${(G.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),R.metadata){if(R.metadata.hasBFloat16)console.log("Supports BFloat16: Yes");if(R.metadata.hasUnifiedMemory)console.log("Unified Memory: Yes")}});else console.log("GPU: Not available");if(console.log(`
+Backend Variant: ${G.variant}`),console.log(`Performance Score: ${G.score}`),G.fit){if(console.log(`
+--- Model Fit Analysis ---`),console.log(`Fits in GPU: ${G.fit.fitsInGpu?"Yes":"No"}`),console.log(`Fits in CPU: ${G.fit.fitsInCpu?"Yes":"No"}`),console.log(`Limiting Factor: ${G.fit.limiting}`),G.limitedFit)console.log(`
+--- Memory-Limited Fit Analysis ---`),console.log(`Limited Total Required: ${(G.limitedFit.totalRequiredBytes/1024/1024/1024).toFixed(2)} GB`),console.log(`Fits in GPU (Limited): ${G.limitedFit.fitsInGpu?"Yes":"No"}`),console.log(`Fits in CPU (Limited): ${G.limitedFit.fitsInCpu?"Yes":"No"}`),console.log(`Limiting Factor (Limited): ${G.limitedFit.limiting}`)}}console.log(`
+=== Full Capabilities JSON ===`),console.log(JSON.stringify(O,null,2)),process.exit(0)}catch(Z){console.error("Failed to get capabilities:",Z.message),process.exit(1)}}async function d2({modelIds:$=[],defaultConfig:X=null}={}){let Z=[];if(console.log(`${g.name} v${g.version}`),console.log(`Generating STT model capabilities comparison...
+`),Z.push(`${g.name} v${g.version}`),Z.push(`## STT Model Capabilities Comparison
+`),!$||$.length===0)console.error("Error: No model IDs provided"),process.exit(1);try{let W=(U={},Q={})=>{let K=Array.isArray(U)?[...U]:{...U};return Object.entries(Q||{}).forEach(([B,L])=>{if(L&&typeof L==="object"&&!Array.isArray(L))K[B]=W(K[B]||{},L);else K[B]=L}),K},j=X||{},{server:V,generators:Y=[],...J}=j,N=(U)=>W(JSON.parse(JSON.stringify(J)),U||{}),H=(U)=>{if(Array.isArray(Y)&&Y.length>0){let Q=Y.filter((K)=>K?.type==="ggml-stt");if(Q.length>0&&U){let K=Q.find((B)=>B.model?.repo_id===U);if(K)return N(K)}}return Object.keys(J).length>0?N({}):null},O=[];for(let U=0;U<$.length;U+=1){let Q=$[U],{repoId:K,filename:B}=p2(Q);console.log(`[${U+1}/${$.length}] Analyzing ${Q}...`);let L=H(K);L={...L||{},model:{...L?.model||{},repo_id:K,...B&&{filename:B}}};let F=await t("ggml-stt",null,{config:L,includeBreakdown:!0});O.push({modelId:Q,repoId:K,filename:B,capabilities:F,modelInfo:F.buttress?.selected||null,modelConfig:F.modelConfig||null})}let q=(U)=>U?(U/1024/1024).toFixed(1):"N/A",_=(U)=>U?"✅":"\uD83D\uDEAB";Z.push("| Model | Size (MB) | Processing Buffer (MB) | Total Required (MB) | Fits GPU | Fits CPU |"),Z.push("|-------|-----------|------------------------|---------------------|----------|----------|"),O.forEach(({modelId:U,modelInfo:Q})=>{let K=q(Q?.modelBytes),B=q(Q?.processingBytes||Q?.kvCacheBytes),L=q(Q?.fit?.totalRequiredBytes),F=_(Q?.fit?.fitsInGpu),E=_(Q?.fit?.fitsInCpu);Z.push(`| ${U} | ${K} | ${B} | ${L} | ${F} | ${E} |`)}),Z.push(`
+---`),Z.push(`
+### System Information`);let G=null;if(process.platform!=="win32")try{G=Y4("uname -a",{encoding:"utf8"}).trim()}catch{}if(G)Z.push(`- **System:** ${G}`);else Z.push(`- **Hostname:** ${m.hostname()}`),Z.push(`- **OS:** ${m.type()} ${m.release()}`);if(Z.push(`- **Platform:** ${process.platform}`),Z.push(`- **CPU Cores:** ${m.cpus().length}`),Z.push(`- **Total System Memory:** ${(m.totalmem()/1024/1024/1024).toFixed(2)} GB`),O.length>0){let Q=O[0].capabilities.buttress?.selected;if(Q){let K=Q.cpuTotalBytes>0?(Q.cpuUsableBytes/Q.cpuTotalBytes*100).toFixed(0):0;if(Z.push(`- **Usable CPU Memory:** ${(Q.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${K}% of ${(Q.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),Q.hasGpu){let B=Q.devices.filter((L)=>L.type==="gpu");if(B.length>0){let L=B[0];Z.push(`- **GPU Backend:** ${L.backend}`),Z.push(`- **GPU Name:** ${L.deviceName}`),Z.push(`- **GPU Total Memory:** ${(L.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let F=Q.gpuTotalBytes>0?(Q.gpuUsableBytes/Q.gpuTotalBytes*100).toFixed(0):0;Z.push(`- **GPU Usable Memory:** ${(Q.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${F}% of ${(Q.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`)}}else Z.push("- **GPU:** Not available")}}Z.push(`
+### Command Used`);let w=process.argv.slice(2).join(" ");if(Z.push(`\`\`\`bash
+${process.argv[0]} ${process.argv[1]} ${w}
+\`\`\``),Z.push(`
+### Package Information`),Z.push(`- **Name:** ${g.name}`),Z.push(`- **Version:** ${g.version}`),g.description)Z.push(`- **Description:** ${g.description}`);if(X&&Object.keys(X).length>0){Z.push(`
+### Configuration`),Z.push("<details>"),Z.push("<summary>Click to expand TOML configuration</summary>"),Z.push("\n```toml");try{let U=g2.stringify(X);Z.push(U)}catch(U){Z.push("# Error serializing config"),Z.push(JSON.stringify(X,null,2))}Z.push("```"),Z.push("</details>")}let A=`ggml-stt-model-capabilities-${new Date().toISOString().replace(/[.:]/g,"-").split("T")[0]}.md`,R=f2.join(process.cwd(),A);u2.writeFileSync(R,Z.join(`
+`),"utf8"),console.log(`
+STT model capabilities table saved to: ${R}`),process.exit(0)}catch(W){console.error("Failed to generate STT model table:",W.message),process.exit(1)}}async function l2({modelId:$=null,defaultConfig:X=null}={}){if(console.log(`${g.name} v${g.version}`),console.log("Testing capabilities for backend: ggml-stt"),$)console.log(`Model: ${$}`);console.log("--------------------------------");try{let{repoId:Z,filename:W}=p2($),j=X||{},{server:V,generators:Y=[],...J}=j,N=(z={},A={})=>{let R=Array.isArray(z)?[...z]:{...z};return Object.entries(A||{}).forEach(([U,Q])=>{if(Q&&typeof Q==="object"&&!Array.isArray(Q))R[U]=N(R[U]||{},Q);else R[U]=Q}),R},H=(z)=>N(JSON.parse(JSON.stringify(J)),z||{}),q=((z)=>{if(Array.isArray(Y)&&Y.length>0){let A=Y.filter((R)=>R?.type==="ggml-stt");if(A.length>0){if(z){let R=A.find((U)=>U.model?.repo_id===z);if(R)return H(R)}}}if(Object.keys(J).length>0)return H({});return null})(Z);if(Z)q={...q||{},model:{...J.runtime,...q?.model||{},repo_id:Z,...W&&{filename:W}}};let _=await t("ggml-stt",null,{config:q,includeBreakdown:!0}),G=_.buttress?.selected||null,w=_.modelConfig||null;if(Z||w?.repoId){console.log(`
+=== Model Information ===`);let z=Z||w?.repoId;if(console.log(`Repository ID: ${z}`),W)console.log(`Filename: ${W}`);if(G?.modelBytes)console.log(`Model Size: ${(G.modelBytes/1024/1024).toFixed(1)} MB`);let A=G?.processingBytes||G?.kvCacheBytes;if(A)console.log(`Processing Buffer: ${(A/1024/1024).toFixed(1)} MB`);if(G?.modelBytes&&A)console.log(`Total Required Memory: ${((G.modelBytes+A)/1024/1024).toFixed(1)} MB`);else if(_.buttress?.selected?.fit){let{totalRequiredBytes:R}=_.buttress.selected.fit;console.log(`Total Required Memory: ${(R/1024/1024).toFixed(1)} MB`)}}if(_.buttress?.selected){let{selected:z}=_.buttress;console.log(`
+=== Hardware Information ===`);let A=null;if(process.platform!=="win32")try{A=Y4("uname -a",{encoding:"utf8"}).trim()}catch{}if(A)console.log(`System: ${A}`);else console.log(`Hostname: ${m.hostname()}`),console.log(`OS: ${m.type()} ${m.release()}`);console.log(`Platform: ${z.platform}`),console.log(`CPU Cores: ${m.cpus().length}`),console.log(`Total System Memory: ${(m.totalmem()/1024/1024/1024).toFixed(2)} GB`);let R=z.cpuTotalBytes>0?(z.cpuUsableBytes/z.cpuTotalBytes*100).toFixed(0):0;if(console.log(`Usable CPU Memory: ${(z.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${R}% of ${(z.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),z.hasGpu)console.log(`
+--- GPU Details ---`),z.devices.filter((Q)=>Q.type==="gpu").forEach((Q)=>{console.log(`GPU Backend: ${Q.backend}`),console.log(`GPU Name: ${Q.deviceName}`),console.log(`GPU Total Memory: ${(Q.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let K=z.gpuTotalBytes>0?(z.gpuUsableBytes/z.gpuTotalBytes*100).toFixed(0):0;if(console.log(`GPU Usable Memory: ${(z.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${K}% of ${(z.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),Q.metadata){if(Q.metadata.hasBFloat16)console.log("Supports BFloat16: Yes");if(Q.metadata.hasUnifiedMemory)console.log("Unified Memory: Yes")}});else console.log("GPU: Not available");if(console.log(`
+Backend Variant: ${z.variant}`),console.log(`Performance Score: ${z.score}`),z.fit)console.log(`
+--- Model Fit Analysis ---`),console.log(`Fits in GPU: ${z.fit.fitsInGpu?"Yes":"No"}`),console.log(`Fits in CPU: ${z.fit.fitsInCpu?"Yes":"No"}`),console.log(`Limiting Factor: ${z.fit.limiting}`)}console.log(`
+=== Full Capabilities JSON ===`),console.log(JSON.stringify(_,null,2)),process.exit(0)}catch(Z){console.error("Failed to get capabilities:",Z.message),process.exit(1)}}var p2=($)=>{if(!$)return{repoId:null,filename:null};let[X,Z]=$.split(":");return{repoId:X,filename:Z||null}};var i2=P(()=>{M0();Z1();y2()});var J4={};i3(J4,{testGgmlSttCapabilities:()=>l2,testGgmlLlmCapabilities:()=>c2,status:()=>V9,startModelDownload:()=>Q4,startGenerator:()=>W9,showSttModelsTable:()=>d2,showModelsTable:()=>m2,globalDownloadManager:()=>W1,ggmlStt:()=>Q9,ggmlLlm:()=>Y9,getModelIdentifier:()=>J9,getCapabilities:()=>t,generatorRegistry:()=>X0,finalizeGenerator:()=>j9});async function W9($,X){let W={"ggml-llm":{create:K2,getId:d4},"ggml-stt":{create:h2,getId:$1}}[$];if(!W)throw Error(`Unsupported backend type: ${$}`);let j=W.getId(X);if(!j)throw Error("Buttress generator config missing repo identifier");let V=`${$}:${j}`,Y=X0.get(V);if(Y)return Y.refCount+=1,Y.instance.resetFinalized?.(),{id:Y.id,info:Y.instance.info};let J=await W.create(V,X,{globalDownloadManager:W1}),N={id:V,type:J.type,instance:J,refCount:1};return X0.set(V,N),{id:V,info:J.info}}async function j9($){let X=X0.get($);if(!X)return!1;if(X.refCount-=1,X.refCount<=0){if(await X.instance.finalize(),!(X.instance.hasPendingReleases?.()??!1))X0.delete($)}return!0}function J9($,X){if($==="ggml-llm")return d4(X);if($==="ggml-stt")return $1(X);return null}async function Q4($,X,Z={}){let j={"ggml-llm":q2,"ggml-stt":I2}[$];if(!j)return{started:!1,localPath:null,repoId:null,error:`Unknown backend type: ${$}`};return j(X,W1,Z)}var X0,W1,Z9=($)=>{let X=X0.get($);if(!X)throw Error(`Unknown generator id "${$}"`);return X},O0=($,X)=>{let Z=Z9($);if(Z.type!==X)throw Error(`Generator "${$}" does not support ${X} backend`);return Z.instance},Y9,Q9,V9;var V4=P(()=>{l4();X1();Z1();s0();i2();X0=new Map,W1={downloads:new Map,getDownload($){return this.downloads.get($)||null},setDownload($,X){this.downloads.set($,X)},deleteDownload($){this.downloads.delete($)},isDownloading($){return this.downloads.has($)},getActiveDownloads(){return Array.from(this.downloads.entries()).map(([$,X])=>({localPath:$,promise:X}))}};Y9={async initContext($,X){return O0($,"ggml-llm").initContext(X)},async completion($,X){return O0($,"ggml-llm").completion(X)},async tokenize($,X){return O0($,"ggml-llm").tokenize(X)},async detokenize($,X){return O0($,"ggml-llm").detokenize(X)},async applyChatTemplate($,X){return O0($,"ggml-llm").applyChatTemplate(X)},async releaseContext($,X){let Z=X0.get($);if(!Z)return{released:!0,alreadyReleased:!0};if(Z.type!=="ggml-llm")throw Error(`Generator "${$}" does not support ggml-llm backend`);return Z.instance.releaseContext(X)}},Q9={async initContext($,X){return O0($,"ggml-stt").initContext(X)},async transcribe($,X){return O0($,"ggml-stt").transcribe(X)},async transcribeData($,X){return O0($,"ggml-stt").transcribeData(X)},async releaseContext($,X){let Z=X0.get($);if(!Z)return{released:!0,alreadyReleased:!0};if(Z.type!=="ggml-stt")throw Error(`Generator "${$}" does not support ggml-stt backend`);return Z.instance.releaseContext(X)}};V9={getFullStatus:()=>d1(X0),getGgmlLlmStatus:()=>C4(X0),getGgmlSttStatus:()=>y4(X0),subscribeToStatus:I4,subscribeToStatusWithId:c1,llmStatusTracker:$0,sttStatusTracker:N0,statusEmitter:e}});import{node as N9}from"@elysiajs/node";import{Elysia as U9}from"elysia";var G9,K0=($)=>new U9({adapter:G9?N9():void 0,...$});var k0=P(()=>{G9=typeof process<"u"&&process.versions&&process.versions.node});import{t as Y0}from"elysia";var H9,O9=({store:{serverInfo:$}})=>({id:$.id,name:$.name,version:$.version,generators:$.generators,authentication:$.authentication}),j1=($)=>{let X=K0(),Z=$.autodiscover.http?.path??"/buttress/info";return X.get(Z,O9,{response:H9}),X};var s2=P(()=>{k0();H9=Y0.Object({id:Y0.String(),name:Y0.String(),version:Y0.String(),generators:Y0.Array(Y0.Object({type:Y0.String()})),authentication:Y0.Object({required:Y0.Boolean(),type:Y0.Literal("device-group")})})});import{t as Q0,file as K9}from"elysia";import{writeFile as n2}from"node:fs/promises";import Y1 from"node:path";var q9,Q1;var r2=P(()=>{k0();q9=typeof process<"u"&&process.versions!=null&&process.versions.node!=null,Q1=K0().post("/buttress/upload",async({body:{file:$},store:{config:X}})=>{let Z=`${Date.now()}-${$.name.replace(/[^\dA-Za-z]/g,"_")}`,W=Y1.join(X.server.temp_file_dir,Z);try{if(q9)await n2(W,await $.stream());else await n2(W,await $.arrayBuffer());return{ok:!0,filename:Z}}catch(j){return{ok:!1,error:String(j)}}},{body:Q0.Object({file:Q0.File()}),response:Q0.Object({ok:Q0.Boolean(),filename:Q0.Optional(Q0.String()),error:Q0.Optional(Q0.String())})}).get("/buttress/download/:filename",async({params:{filename:$},store:{config:X},status:Z})=>{let W=Y1.join(X.server.temp_file_dir,$);if(Y1.relative(X.server.temp_file_dir,W).includes(".."))return Z(400),"Invalid file path";return K9(W)},{params:Q0.Object({filename:Q0.String()})})});import J1 from"node:path";import t2 from"node:fs/promises";import{fileURLToPath as _9}from"node:url";var o2,L9=async()=>{let $=[J1.join(o2,"..","public","status.html"),J1.join(o2,"..","..","public","status.html")];return(await Promise.all($.map((Z)=>t2.access(Z).then(()=>Z,()=>null)))).find((Z)=>Z!==null)??null},A9=($)=>{let{status:X}=$;if(X?.getFullStatus)return X.getFullStatus();return{timestamp:new Date().toISOString(),ggmlLlm:{generators:[],history:{}},ggmlStt:{generators:[],history:{}}}},a2=async()=>{let $=await L9();if(!$)return console.error("[Status] Failed to find status.html in candidate paths"),new Response("Status page not found",{status:404,headers:{"Content-Type":"text/plain"}});try{let X=await t2.readFile($,"utf-8");return new Response(X,{headers:{"Content-Type":"text/html; charset=utf-8"}})}catch(X){return console.error("[Status] Failed to serve status page:",X),new Response("Status page not found",{status:404,headers:{"Content-Type":"text/plain"}})}},V1;var e2=P(()=>{k0();o2=J1.dirname(_9(import.meta.url)),V1=K0().get("/status",a2).get("/status/",a2).get("/buttress/status",({store:{backend:$}})=>A9($))});import{t as T,sse as N4}from"elysia";import{cors as z9}from"@elysiajs/cors";async function R9($,X,Z){let j=(X.generators||[]).filter((z)=>z.type==="ggml-llm");if(j.length===0)throw Error('No ggml-llm generator configured. Add a [[generators]] with type = "ggml-llm" to your config.');let V=j[0],Y=Z||V.model?.repo_id;if(Z){let z=j.find((A)=>A.model?.repo_id===Z);if(z)V=z}else Y=V.model?.repo_id;let J=Y,N=$3.get(J);if(N?.initialized)return N;let{generators:H,server:O,...q}=X.global||{},_={...q,...V,model:{...V.model,repo_id:Y}};console.log(`[OpenAI] Creating generator for ${J}`);let{id:G}=await $.startGenerator("ggml-llm",_),w={id:G,config:_,repoId:Y,initialized:!1};return $3.set(J,w),await $.ggmlLlm.initContext(G,{}),w.initialized=!0,console.log(`[OpenAI] Generator ready: ${J}`),w}async function w9($,X,Z,W){let j=$.getReader(),V="",Y=null,J=null,N="stop",H={prompt_tokens:0,completion_tokens:0,total_tokens:0};try{let q=!1;while(!q){let _=await j.read();if({done:q}=_,q)break;let{event:G,data:w}=_.value;if(G==="token"){if(w.content!=null)V+=w.content;else if(w.token!=null)V+=w.token}else if(G==="result"){if(w.text)V=w.text;else if(w.content)V=w.content;if(w.reasoning_content)Y=w.reasoning_content;if(w.tool_calls?.length>0)J=w.tool_calls.map((z,A)=>({id:z.id||`call_${X}_${A}`,type:"function",function:{name:z.function?.name||"",arguments:z.function?.arguments||""}})),N="tool_calls";else N=w.interrupted?"length":"stop";H={prompt_tokens:w.prompt_tokens||w.promptTokens||0,completion_tokens:w.tokens_predicted||w.tokensPredicted||0,total_tokens:(w.prompt_tokens||w.promptTokens||0)+(w.tokens_predicted||w.tokensPredicted||0)}}else if(G==="error")throw Error(w.message)}}finally{j.cancel().catch(()=>{})}let O={role:"assistant",content:V||null};if(Y)O.reasoning_content=Y;if(J)O.tool_calls=J;return{id:X,object:"chat.completion",created:Z,model:W,choices:[{index:0,message:O,finish_reason:N}],usage:H}}function U4({global:$}){let X=K0({prefix:"/oai-compat"});return X.use(z9({origin:$?.openai_compat?.cors_allowed_origins??!1,methods:["GET","POST","OPTIONS"],allowedHeaders:["Content-Type","Authorization"],maxAge:86400,preflight:!0})),X.get("/v1/models",({store:Z})=>{let{config:W}=Z,Y=(W.generators||[]).filter((J)=>J.type==="ggml-llm").map((J)=>{return{id:J.model?.repo_id||"ggml-llm",object:"model",created:Math.floor(Date.now()/1000),owned_by:"local"}});if(Y.length===0)Y.push({id:"ggml-llm",object:"model",created:Math.floor(Date.now()/1000),owned_by:"local"});return{object:"list",data:Y}}),X.post("/v1/chat/completions",async function*({body:W,set:j,store:V}){let{config:Y,backend:J}=V,{messages:N=[],stream:H=!1,model:O,tools:q,temperature:_,stop:G,top_p:w,max_tokens:z,presence_penalty:A,frequency_penalty:R,tool_choice:U,stream_options:Q}=W;if(!N||N.length===0)return j.status=400,{error:{message:"messages is required and must not be empty",type:"invalid_request_error"}};try{let K=await R9(J,Y,O),B=B9(),L=Math.floor(Date.now()/1000),F=K.repoId||"ggml-llm",E={reasoning_format:"auto",messages:N,jinja:!0,add_generation_prompt:!0};if(_!=null)E.temperature=_;if(w!=null)E.top_p=w;if(z!=null)E.n_predict=z;if(G!=null)E.stop=Array.isArray(G)?G:[G];if(A!=null)E.presence_penalty=A;if(R!=null)E.frequency_penalty=R;if(q!=null)E.tools=q;if(U!=null)E.tool_choice=U;E.enable_thinking=!1;let M=await J.ggmlLlm.completion(K.id,{options:E});if(!H)return await w9(M,B,L,F);let x=Q?.include_usage===!0,D=M.getReader(),I="",b="",u=new Map,r=new Map;try{let k=!1;while(!k){let d=await D.read();if({done:k}=d,k)break;let{event:v,data:S}=d.value;if(v==="token"){let C={};if(S.content!=null){let h=S.content;if(h.length>I.length)C.content=h.slice(I.length),I=h}if(S.reasoning_content!=null){let h=S.reasoning_content;if(h.length>b.length)C.reasoning_content=h.slice(b.length),b=h}if(S.tool_calls?.length>0){let h=[];if(S.tool_calls.forEach((p,f)=>{let s={index:f};if(!r.has(f))r.set(f,p.id||`call_${B}_${f}`),s.id=r.get(f),s.type="function";let W0=p.function?.arguments||"",A0=u.get(f)||"",l={};if(!u.has(f)&&p.function?.name)l.name=p.function.name;if(W0.length>A0.length)l.arguments=W0.slice(A0.length),u.set(f,W0);if(Object.keys(l).length>0)s.function=l,h.push(s);else if(s.id)s.function={name:p.function?.name||"",arguments:""},h.push(s)}),h.length>0)C.tool_calls=h}if(Object.keys(C).length>0)yield N4({data:JSON.stringify({id:B,object:"chat.completion.chunk",created:L,model:F,choices:[{index:0,delta:C,finish_reason:null}]})})}else if(v==="result"){let C="stop";if(S.tool_calls?.length>0||r.size>0)C="tool_calls";else if(S.interrupted)C="length";let h={id:B,object:"chat.completion.chunk",created:L,model:F,choices:[{index:0,delta:{},finish_reason:C}]};if(x)h.usage={prompt_tokens:S.prompt_tokens||S.promptTokens||0,completion_tokens:S.tokens_predicted||S.tokensPredicted||0,total_tokens:(S.prompt_tokens||S.promptTokens||0)+(S.tokens_predicted||S.tokensPredicted||0)};yield N4({data:JSON.stringify(h)})}else if(v==="error")yield N4({data:JSON.stringify({error:{message:S.message,type:"server_error"}})})}yield N4({data:"[DONE]"})}finally{D.cancel().catch(()=>{})}}catch(K){return console.error("[OpenAI] Chat completion error:",K),j.status=500,{error:{message:K.message||"Internal server error",type:"server_error"}}}},{body:T.Object({model:T.Optional(T.String()),messages:T.Array(T.Any()),stream:T.Optional(T.Boolean()),temperature:T.Optional(T.Number()),top_p:T.Optional(T.Number()),max_tokens:T.Optional(T.Number()),stop:T.Optional(T.Union([T.String(),T.Array(T.String())])),presence_penalty:T.Optional(T.Number()),frequency_penalty:T.Optional(T.Number()),tools:T.Optional(T.Array(T.Any())),tool_choice:T.Optional(T.Any()),stream_options:T.Optional(T.Object({include_usage:T.Optional(T.Boolean())}))})}),X}var B9=()=>`chatcmpl-${Date.now()}-${Math.random().toString(36).slice(2,9)}`,$3;var X3=P(()=>{k0();$3=new Map});var Z3=P(()=>{s2();r2();e2();X3()});import E9 from"node:os";import M9 from"node:path";import x9 from"node-machine-id";import S9 from"bytes";import P9 from"ms";var D0=($={},X={})=>{let Z=Array.isArray($)?[...$]:{...$};return Object.entries(X||{}).forEach(([W,j])=>{if(j&&typeof j==="object"&&!Array.isArray(j))Z[W]=D0(Z[W]||{},j);else Z[W]=j}),Z},w0=($)=>{if(!$)return null;if(typeof $==="object")return JSON.parse(JSON.stringify($));return null},N1=($,X)=>{let Z=w0($)||{},W=w0(X)||{};return D0(Z,W)},W3=($,X)=>D0(JSON.parse(JSON.stringify($.global)),X||{}),U1=($,X,Z,W)=>{if($.generators.length>0){let j=$.generators.filter((V)=>V?.type===Z);if(j.length>0){if(W){let V=j.find((Y)=>X.getModelIdentifier(Z,Y)===W);if(V)return W3($,V)}}}if(Object.keys($.global).length>0)return W3($,{});return null},j3,T9=($)=>{if(!$)return null;if($===!0)return{...j3};return D0(j3,$)},V3=($,X)=>{if(!$.generators||$.generators.length===0)return X.map((W)=>({type:W}));let Z=new Set;if($.generators.forEach((W)=>{if(W.type)Z.add(W.type)}),Z.size===0)return X.map((W)=>({type:W}));return Array.from(Z).map((W)=>({type:W}))},Y3=($,X,Z)=>{if($===void 0)return Z;if(typeof $==="number")return $;return X($)??Z},k9=2080,Q3=60000,J3=52428800,G1=($)=>{let X=x9.machineIdSync(),Z={server:{id:`buttress-${X}`,name:`Buttress Server (${X.slice(-8)})`,port:k9,temp_file_dir:M9.join(E9.tmpdir(),".buttress"),session_timeout:Q3,max_body_size:J3},autodiscover:!1},W=D0(Z,w0($)||{}),j=Array.isArray(W.generators)?W.generators:[],{server:V,generators:Y,autodiscover:J,...N}=W;return{autodiscover:T9(J),server:{id:V.id,name:V.name,port:V.port,log_level:V.log_level,temp_file_dir:V.temp_file_dir,max_body_size:Y3(V.max_body_size,S9.parse,J3),session_timeout:Y3(V.session_timeout,P9,Q3)},global:N,generators:j}};var b0=P(()=>{j3={udp:{port:8089,announcements:{enabled:!0,interval:5000},requests:{enabled:!0,responseDelay:100}},http:{enabled:!0,path:"/buttress/info",cors:!0}}});import{z as Z0}from"zod";var N3,U3;var G3=P(()=>{b0();N3={getCapabilities:Z0.tuple([Z0.object({type:Z0.string().optional().default("ggml-llm"),config:Z0.any().optional(),currentClientCapabilities:Z0.any().optional(),options:Z0.any().optional()}).nullable().optional()]),startGenerator:Z0.tuple([Z0.string(),Z0.any().optional()]),finalizeGenerator:Z0.tuple([Z0.string()])},U3={async getCapabilities({backend:$,config:X},Z=null){console.log("[Server] Get Capabilities:",Z);let W=Z||{type:"ggml-llm"},{type:j="ggml-llm",config:V,currentClientCapabilities:Y=null,options:J={}}=W,N=w0(V),H=$.getModelIdentifier(j,N),O=U1(X,$,j,H),q=N1(O,V);if(Object.keys(q).length===0)throw Error("Buttress server missing generator configuration");if(q.backend=q.backend||{},!q.backend.type)q.backend.type=j;return $.getCapabilities(j,Y,{...J,config:q})},async startGenerator({backend:$,config:X,session:Z},W,j){console.log("[Server] Start Generator:",W,j);let V=w0(j),Y=$.getModelIdentifier(W,V),J=U1(X,$,W,Y),N=N1(J,j);if(Object.keys(N).length===0)throw Error("Buttress server missing generator configuration");if(N.backend=N.backend||{},!N.backend.type)N.backend.type=W;let H=await $.startGenerator(W,N);return Z.generators.add(H.id),H},async finalizeGenerator({backend:$,session:X},Z){return console.log("[Server] Finalize Generator:",Z),X.generators.delete(Z),$.finalizeGenerator(Z)}}});import{z as i}from"zod";import{ReadableStream as D9}from"node:stream/web";var H3,O3;var K3=P(()=>{H3={initContext:i.tuple([i.string(),i.any().optional()]),completion:i.tuple([i.string(),i.any().optional()]),tokenize:i.tuple([i.string(),i.any()]),detokenize:i.tuple([i.string(),i.any()]),applyChatTemplate:i.tuple([i.string(),i.any()]),releaseContext:i.tuple([i.string()])},O3={initContext({backend:$,session:X},Z,W){return new D9({async start(j){try{let V=await $.ggmlLlm.initContext(Z,{...W,onProgress:(N)=>{j.enqueue({event:"progress",data:{progress:N}})}});X.initializedContexts.add(Z),await new Promise((N)=>setTimeout(N,1000));let{download:Y,...J}=V||{};j.enqueue({event:"result",data:{result:J}}),j.close()}catch(V){j.error(V)}}})},completion({backend:$},X,Z){return console.log("[Server] Completion:",{id:X,property:Z}),$.ggmlLlm.completion(X,Z)},async tokenize({backend:$},X,Z){return console.log("[Server] Tokenize:",{id:X,property:Z}),$.ggmlLlm.tokenize(X,Z)},async detokenize({backend:$},X,Z){return console.log("[Server] Detokenize:",{id:X,property:Z}),$.ggmlLlm.detokenize(X,Z)},async applyChatTemplate({backend:$},X,Z){return console.log("[Server] Apply Chat Template:",{id:X,property:Z}),$.ggmlLlm.applyChatTemplate(X,Z)},async releaseContext({backend:$,session:X},Z,W){if(console.log("[Server] Release Context:",{id:Z,force:W}),!X.initializedContexts.has(Z))return console.log("[Server] Release Context skipped - not initialized by this session:",{id:Z}),{released:!1,skipped:!0};return X.initializedContexts.delete(Z),$.ggmlLlm.releaseContext(Z,{force:W})}}});import{z as a}from"zod";import{ReadableStream as b9}from"node:stream/web";import v9 from"node:path";var q3,_3;var L3=P(()=>{q3={initContext:a.tuple([a.string(),a.any().optional()]),transcribe:a.tuple([a.string(),a.string(),a.any().optional()]),transcribeData:a.tuple([a.string(),a.union([a.instanceof(Buffer),a.instanceof(Uint8Array)]),a.any().optional()]),releaseContext:a.tuple([a.string()])},_3={initContext({backend:$,session:X},Z,W){return new b9({async start(j){try{let V=await $.ggmlStt.initContext(Z,{...W,onProgress:(N)=>{j.enqueue({event:"progress",data:{progress:N}})}});X.initializedContexts.add(Z),await new Promise((N)=>setTimeout(N,1000));let{download:Y,...J}=V||{};j.enqueue({event:"result",data:{result:J}}),j.close()}catch(V){j.error(V)}}})},async transcribe({backend:$,config:{server:X}},Z,W,j){return console.log("[Server] Transcribe:",{id:Z,audioPath:W,options:j}),$.ggmlStt.transcribe(Z,{audioPath:v9.join(X.temp_file_dir,W),options:j})},async transcribeData({backend:$},X,Z,W){return console.log("[Server] Transcribe Data:",{id:X,audioDataLength:Z?.length||0,options:W}),$.ggmlStt.transcribeData(X,{audioData:Z,options:W})},async releaseContext({backend:$,session:X},Z,W){if(console.log("[Server] Release STT Context:",{id:Z,force:W}),!X.initializedContexts.has(Z))return console.log("[Server] Release STT Context skipped - not initialized by this session:",{id:Z}),{released:!1,skipped:!0};return X.initializedContexts.delete(Z),$.ggmlStt.releaseContext(Z,{force:W})}}});var h9,A3,z3;var B3=P(()=>{G3();K3();L3();h9={common:U3,ggmlLlm:O3,ggmlStt:_3},A3={common:N3,ggmlLlm:H3,ggmlStt:q3},z3=h9});import{Buffer as G4}from"node:buffer";var R3=($)=>{try{return JSON.parse($,(X,Z)=>{if(!Z)return Z;if(Z?.type==="Buffer"&&Z?.data)return G4.from(Z.data,"base64");if(Z?.type==="Uint8Array"&&Z?.data){let W=G4.from(Z.data,"base64");return W.buffer.slice(W.byteOffset,W.byteOffset+W.byteLength)}if(Z?.type==="Error"&&Z?.name&&Z?.message)return Error(Z.name,Z.message);return Z})}catch{return $}},H4=($)=>{try{return JSON.stringify($,(X,Z)=>{if(Z instanceof Error)return{type:"Error",name:Z.name,message:Z.message};if(Z instanceof G4)return{type:"Buffer",data:Z.toString("base64")};if(Z instanceof Uint8Array)return{type:"Uint8Array",data:G4.from(Z).toString("base64")};return Z})}catch{return $}};var w3=()=>{};var H1="1.0",F3=8089;import I9 from"node:dgram";class O1{name="udp";socket=null;announcementTimer=null;config;getServerInfo;port;constructor($,X){this.config=$,this.getServerInfo=X,this.port=$.port??F3}async start(){if(this.socket=I9.createSocket({type:"udp4",reuseAddr:!0}),this.socket.on("message",($,X)=>{this.handleMessage($,X)}),this.socket.on("error",($)=>{console.error("[Autodiscover UDP] Socket error:",$.message)}),await new Promise(($,X)=>{this.socket.bind(this.port,()=>{this.socket.setBroadcast(!0),console.log(`[Autodiscover UDP] Listening on port ${this.port}`),$()}),this.socket.once("error",X)}),this.config.announcements.enabled){let $=this.config.announcements.interval??5000;this.announcementTimer=setInterval(()=>{this.sendAnnouncement()},$),this.sendAnnouncement()}}async stop(){if(this.announcementTimer)clearInterval(this.announcementTimer),this.announcementTimer=null;if(this.socket)await new Promise(($)=>{this.socket.close(()=>$())}),this.socket=null}handleMessage($,X){try{let Z=JSON.parse($.toString());if(Z.t==="QUERY"&&this.config.requests.enabled){let W=Z.d,j=this.config.requests.responseDelay??0,V=j>0?Math.random()*j:0;setTimeout(()=>{this.sendResponse(W.id,X)},V)}}catch{}}sendAnnouncement(){if(!this.socket)return;let $=this.getServerInfo(),X={t:"ANNOUNCE",v:H1,d:{info:$}},Z=Buffer.from(JSON.stringify(X));this.socket.send(Z,0,Z.length,this.port,"255.255.255.255",(W)=>{if(W)console.error("[Autodiscover UDP] Announcement error:",W.message)})}sendResponse($,X){if(!this.socket)return;let Z=this.getServerInfo(),W={t:"RESPONSE",v:H1,d:{request_id:$,info:Z}},j=Buffer.from(JSON.stringify(W));this.socket.send(j,0,j.length,X.port,X.address,(V)=>{if(V)console.error("[Autodiscover UDP] Response error:",V.message)})}}var E3=()=>{};class K1{config;getServerInfo;transports=[];started=!1;constructor($,X){this.config=$;this.getServerInfo=X;if($.udp?.announcements?.enabled||$.udp?.requests?.enabled)this.transports.push(new O1($.udp,X))}async start(){if(this.started)return;(await Promise.allSettled(this.transports.map((X)=>X.start()))).forEach((X,Z)=>{if(X.status==="rejected")console.error(`[Autodiscover] Failed to start ${this.transports[Z].name}:`,X.reason)}),this.started=!0}async stop(){if(!this.started)return;await Promise.allSettled(this.transports.map(($)=>$.stop())),this.started=!1}}var M3=P(()=>{E3()});import C9 from"node:os";var O4=()=>{let $=C9.networkInterfaces();return Object.values($).flat().find((Z)=>Z?.family==="IPv4"&&!Z?.internal)?.address||null};var q1=()=>{};import c from"node:os";import x3 from"node:fs";import S3 from"node:path";import{execSync as K4}from"node:child_process";import P3 from"@iarna/toml";async function _1({modelIds:$=[],defaultConfig:X=null}={}){let Z=[];if(console.log(`${"@fugood/buttress-server"} v${"2.23.0-beta.53"}`),console.log(`Generating model capabilities comparison...
+`),Z.push(`${"@fugood/buttress-server"} v${"2.23.0-beta.53"}`),Z.push(`## Model Capabilities Comparison
+`),!$||$.length===0)console.error("Error: No model IDs provided"),process.exit(1);try{let W=(U={},Q={})=>{let K=Array.isArray(U)?[...U]:{...U};return Object.entries(Q||{}).forEach(([B,L])=>{if(L&&typeof L==="object"&&!Array.isArray(L))K[B]=W(K[B]||{},L);else K[B]=L}),K},j=X||{},{server:V,generators:Y=[],...J}=j,N=(U)=>W(JSON.parse(JSON.stringify(J)),U||{}),H=(U)=>{if(Array.isArray(Y)&&Y.length>0){let Q=Y.filter((K)=>K?.type==="ggml-llm");if(Q.length>0&&U){let K=Q.find((B)=>B.model?.repo_id===U);if(K)return N(K)}}return Object.keys(J).length>0?N({}):null},O=[];for(let U=0;U<$.length;U+=1){let Q=$[U];console.log(`[${U+1}/${$.length}] Analyzing ${Q}...`);let K=H(Q);K={...K||{},model:{...J.runtime,...K?.model||{},repo_id:Q}};let B=await t("ggml-llm",null,{config:K,includeBreakdown:!0});O.push({modelId:Q,capabilities:B,modelInfo:B.buttress?.selected||null,modelConfig:B.modelConfig||null})}let q=(U)=>U?(U/1024/1024/1024).toFixed(2):"N/A",_=(U)=>U?"✅":"\uD83D\uDEAB";Z.push("| Model ID | Size (GB) | Context Size | KV Cache Size (GB) | Recurrent Mem (GB) | Total Required Memory (GB) | Fits GPU (Full) | Fits CPU (Full) |"),Z.push("|----------|-----------|--------------|--------------------|--------------------|----------------------------|-----------------|-----------------|"),O.forEach(({modelId:U,modelInfo:Q,modelConfig:K})=>{let B=q(Q?.modelBytes),L=K?.nCtx||Q?.kvInfo?.nCtxTrain||"N/A",F=G0(Q),E=Number(L),M=Q?.kvCacheBytes||(F&&Number.isFinite(E)&&E>0?F(E):F&&F(Q?.kvInfo?.nCtxTrain||0))||null,x=q(M),D=Q?.recurrentMemoryBytes||0,I=D>0?q(D):"-",b=q(Q?.modelBytes&&(M!=null||D>0)?Q.modelBytes+(M||0)+D:Q?.fit?.totalRequiredBytes),u=_(Q?.fit?.fitsInGpu),r=_(Q?.fit?.fitsInCpu);Z.push(`| ${U} | ${B} | ${L} | ${x} | ${I} | ${b} | ${u} | ${r} |`);let k=Q?.memoryLimitedCtx!=null||Q?.limitedFit!=null,d=!Q?.fit?.fitsInGpu||!Q?.fit?.fitsInCpu;if(k&&d){let v=Q?.memoryLimitedCtx||L,S=Number(v),C=Q?.limitedKvCacheBytes||F&&Number.isFinite(S)&&S>0&&F(S)||null,h=q(C),p=q(Q?.modelBytes&&(C!=null||D>0)?Q.modelBytes+(C||0)+D:Q?.limitedFit?.totalRequiredBytes),f=_(Q?.limitedFit?.fitsInGpu),s=_(Q?.limitedFit?.fitsInCpu);if(v!==L||h!==x||p!==b)Z.push(`| ↳ Limited | ${B} | ${v} | ${h} | ${I} | ${p} | ${f} | ${s} |`)}}),Z.push(`
+---`),Z.push(`
+### System Information`);let G=null;if(process.platform!=="win32")try{G=K4("uname -a",{encoding:"utf8"}).trim()}catch{}if(G)Z.push(`- **System:** ${G}`);else Z.push(`- **Hostname:** ${c.hostname()}`),Z.push(`- **OS:** ${c.type()} ${c.release()}`);if(Z.push(`- **Platform:** ${process.platform}`),Z.push(`- **CPU Cores:** ${c.cpus().length}`),Z.push(`- **Total System Memory:** ${(c.totalmem()/1024/1024/1024).toFixed(2)} GB`),O.length>0){let Q=O[0].capabilities.buttress?.selected;if(Q){let K=Q.cpuTotalBytes>0?(Q.cpuUsableBytes/Q.cpuTotalBytes*100).toFixed(0):0;if(Z.push(`- **Usable CPU Memory:** ${(Q.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${K}% of ${(Q.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),Q.hasGpu){let B=Q.devices.filter((L)=>L.type==="gpu");if(B.length>0){let L=B[0];Z.push(`- **GPU Backend:** ${L.backend}`),Z.push(`- **GPU Name:** ${L.deviceName}`),Z.push(`- **GPU Total Memory:** ${(L.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let F=Q.gpuTotalBytes>0?(Q.gpuUsableBytes/Q.gpuTotalBytes*100).toFixed(0):0;Z.push(`- **GPU Usable Memory:** ${(Q.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${F}% of ${(Q.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`)}}else Z.push("- **GPU:** Not available")}}Z.push(`
+### Command Used`);let w=process.argv.slice(2).join(" ");if(Z.push(`\`\`\`bash
+${process.argv[0]} ${process.argv[1]} ${w}
+\`\`\``),Z.push(`
+### Package Information`),Z.push(`- **Name:** ${"@fugood/buttress-server"}`),Z.push(`- **Version:** ${"2.23.0-beta.53"}`),X&&Object.keys(X).length>0){Z.push(`
+### Configuration`),Z.push("<details>"),Z.push("<summary>Click to expand TOML configuration</summary>"),Z.push("\n```toml");try{let U=P3.stringify(X);Z.push(U)}catch(U){Z.push("# Error serializing config"),Z.push(JSON.stringify(X,null,2))}Z.push("```"),Z.push("</details>")}let A=`ggml-llm-model-capabilities-${new Date().toISOString().replace(/[.:]/g,"-").split("T")[0]}.md`,R=S3.join(process.cwd(),A);x3.writeFileSync(R,Z.join(`
+`),"utf8"),console.log(`
+Model capabilities table saved to: ${R}`),process.exit(0)}catch(W){console.error("Failed to generate model table:",W.message),process.exit(1)}}async function k3({modelId:$=null,defaultConfig:X=null}={}){if(console.log(`${"@fugood/buttress-server"} v${"2.23.0-beta.53"}`),console.log("Testing capabilities for backend: ggml-llm"),$)console.log(`Model: ${$}`);console.log("--------------------------------");try{let Z=X||{},{server:W,generators:j=[],...V}=Z,Y=(G={},w={})=>{let z=Array.isArray(G)?[...G]:{...G};return Object.entries(w||{}).forEach(([A,R])=>{if(R&&typeof R==="object"&&!Array.isArray(R))z[A]=Y(z[A]||{},R);else z[A]=R}),z},J=(G)=>Y(JSON.parse(JSON.stringify(V)),G||{}),H=((G)=>{if(Array.isArray(j)&&j.length>0){let w=j.filter((z)=>z?.type==="ggml-llm");if(w.length>0){if(G){let z=w.find((A)=>A.model?.repo_id===G);if(z)return J(z)}}}if(Object.keys(V).length>0)return J({});return null})($);if($)H={...H||{},model:{...H?.model||{},repo_id:$}};let O=await t("ggml-llm",null,{config:H,includeBreakdown:!0}),q=O.buttress?.selected||null,_=O.modelConfig||null;if($||_?.repoId){console.log(`
+=== Model Information ===`);let G=$||_?.repoId;if(console.log(`Repository ID: ${G}`),_?.quantization)console.log(`Quantization: ${_.quantization}`);if(_?.nCtx)console.log(`Context Length: ${_.nCtx}`);if(q?.quantization){let{fileType:A}=q.quantization;if(A!=null)console.log(`Model File Type (GGUF): ${A}`)}let w=_?.cache_type_k||"f16",z=_?.cache_type_v||"f16";if(console.log(`KV Cache Type: K=${w}, V=${z}`),q?.modelBytes&&q?.kvCacheBytes!=null){if(console.log(`Model Size: ${(q.modelBytes/1024/1024/1024).toFixed(2)} GB`),q.kvInfo)console.log(`KV Cache Size: ${(q.kvCacheBytes/1024/1024/1024).toFixed(2)} GB (KV info: ${JSON.stringify(q.kvInfo)})`);else console.log(`KV Cache Size: ${(q.kvCacheBytes/1024/1024/1024).toFixed(2)} GB`);let A=q.recurrentMemoryBytes||0;if(A>0)console.log(`Recurrent Memory: ${(A/1024/1024/1024).toFixed(2)} GB`);let R=q.modelBytes+q.kvCacheBytes+A;if(console.log(`Total Required Memory: ${(R/1024/1024/1024).toFixed(2)} GB`),q.memoryLimitedCtx!=null){let U=q.memoryLimitedCtx,Q=q.kvInfo?.nCtxTrain;if(Q)console.log(`
+Memory-Limited Context: ${U} (Train: ${Q})`);else console.log(`
+Memory-Limited Context: ${U}`);if(q.limitedKvCacheBytes!=null)console.log(`Limited KV Cache Size: ${(q.limitedKvCacheBytes/1024/1024/1024).toFixed(2)} GB`)}}else if(O.buttress?.selected?.fit){let{totalRequiredBytes:A}=O.buttress.selected.fit;console.log(`Total Required Memory: ${(A/1024/1024/1024).toFixed(2)} GB`)}}if(O.buttress?.selected){let{selected:G}=O.buttress;console.log(`
+=== Hardware Information ===`);let w=null;if(process.platform!=="win32")try{w=K4("uname -a",{encoding:"utf8"}).trim()}catch{}if(w)console.log(`System: ${w}`);else console.log(`Hostname: ${c.hostname()}`),console.log(`OS: ${c.type()} ${c.release()}`);console.log(`Platform: ${G.platform}`),console.log(`CPU Cores: ${c.cpus().length}`),console.log(`Total System Memory: ${(c.totalmem()/1024/1024/1024).toFixed(2)} GB`);let z=G.cpuTotalBytes>0?(G.cpuUsableBytes/G.cpuTotalBytes*100).toFixed(0):0;if(console.log(`Usable CPU Memory: ${(G.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${z}% of ${(G.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),G.hasGpu)console.log(`
+--- GPU Details ---`),G.devices.filter((R)=>R.type==="gpu").forEach((R)=>{console.log(`GPU Backend: ${R.backend}`),console.log(`GPU Name: ${R.deviceName}`),console.log(`GPU Total Memory: ${(R.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let U=G.gpuTotalBytes>0?(G.gpuUsableBytes/G.gpuTotalBytes*100).toFixed(0):0;if(console.log(`GPU Usable Memory: ${(G.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${U}% of ${(G.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),R.metadata){if(R.metadata.hasBFloat16)console.log("Supports BFloat16: Yes");if(R.metadata.hasUnifiedMemory)console.log("Unified Memory: Yes")}});else console.log("GPU: Not available");if(console.log(`
+Backend Variant: ${G.variant}`),console.log(`Performance Score: ${G.score}`),G.fit){if(console.log(`
+--- Model Fit Analysis ---`),console.log(`Fits in GPU: ${G.fit.fitsInGpu?"Yes":"No"}`),console.log(`Fits in CPU: ${G.fit.fitsInCpu?"Yes":"No"}`),console.log(`Limiting Factor: ${G.fit.limiting}`),G.limitedFit)console.log(`
+--- Memory-Limited Fit Analysis ---`),console.log(`Limited Total Required: ${(G.limitedFit.totalRequiredBytes/1024/1024/1024).toFixed(2)} GB`),console.log(`Fits in GPU (Limited): ${G.limitedFit.fitsInGpu?"Yes":"No"}`),console.log(`Fits in CPU (Limited): ${G.limitedFit.fitsInCpu?"Yes":"No"}`),console.log(`Limiting Factor (Limited): ${G.limitedFit.limiting}`)}}console.log(`
+=== Full Capabilities JSON ===`),console.log(JSON.stringify(O,null,2)),process.exit(0)}catch(Z){console.error("Failed to get capabilities:",Z.message),process.exit(1)}}async function L1({modelIds:$=[],defaultConfig:X=null}={}){let Z=[];if(console.log(`${"@fugood/buttress-server"} v${"2.23.0-beta.53"}`),console.log(`Generating STT model capabilities comparison...
+`),Z.push(`${"@fugood/buttress-server"} v${"2.23.0-beta.53"}`),Z.push(`## STT Model Capabilities Comparison
+`),!$||$.length===0)console.error("Error: No model IDs provided"),process.exit(1);try{let W=(U={},Q={})=>{let K=Array.isArray(U)?[...U]:{...U};return Object.entries(Q||{}).forEach(([B,L])=>{if(L&&typeof L==="object"&&!Array.isArray(L))K[B]=W(K[B]||{},L);else K[B]=L}),K},j=X||{},{server:V,generators:Y=[],...J}=j,N=(U)=>W(JSON.parse(JSON.stringify(J)),U||{}),H=(U)=>{if(Array.isArray(Y)&&Y.length>0){let Q=Y.filter((K)=>K?.type==="ggml-stt");if(Q.length>0&&U){let K=Q.find((B)=>B.model?.repo_id===U);if(K)return N(K)}}return Object.keys(J).length>0?N({}):null},O=[];for(let U=0;U<$.length;U+=1){let Q=$[U],{repoId:K,filename:B}=T3(Q);console.log(`[${U+1}/${$.length}] Analyzing ${Q}...`);let L=H(K);L={...L||{},model:{...L?.model||{},repo_id:K,...B&&{filename:B}}};let F=await t("ggml-stt",null,{config:L,includeBreakdown:!0});O.push({modelId:Q,repoId:K,filename:B,capabilities:F,modelInfo:F.buttress?.selected||null,modelConfig:F.modelConfig||null})}let q=(U)=>U?(U/1024/1024).toFixed(1):"N/A",_=(U)=>U?"✅":"\uD83D\uDEAB";Z.push("| Model | Size (MB) | Processing Buffer (MB) | Total Required (MB) | Fits GPU | Fits CPU |"),Z.push("|-------|-----------|------------------------|---------------------|----------|----------|"),O.forEach(({modelId:U,modelInfo:Q})=>{let K=q(Q?.modelBytes),B=q(Q?.processingBytes||Q?.kvCacheBytes),L=q(Q?.fit?.totalRequiredBytes),F=_(Q?.fit?.fitsInGpu),E=_(Q?.fit?.fitsInCpu);Z.push(`| ${U} | ${K} | ${B} | ${L} | ${F} | ${E} |`)}),Z.push(`
+---`),Z.push(`
+### System Information`);let G=null;if(process.platform!=="win32")try{G=K4("uname -a",{encoding:"utf8"}).trim()}catch{}if(G)Z.push(`- **System:** ${G}`);else Z.push(`- **Hostname:** ${c.hostname()}`),Z.push(`- **OS:** ${c.type()} ${c.release()}`);if(Z.push(`- **Platform:** ${process.platform}`),Z.push(`- **CPU Cores:** ${c.cpus().length}`),Z.push(`- **Total System Memory:** ${(c.totalmem()/1024/1024/1024).toFixed(2)} GB`),O.length>0){let Q=O[0].capabilities.buttress?.selected;if(Q){let K=Q.cpuTotalBytes>0?(Q.cpuUsableBytes/Q.cpuTotalBytes*100).toFixed(0):0;if(Z.push(`- **Usable CPU Memory:** ${(Q.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${K}% of ${(Q.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),Q.hasGpu){let B=Q.devices.filter((L)=>L.type==="gpu");if(B.length>0){let L=B[0];Z.push(`- **GPU Backend:** ${L.backend}`),Z.push(`- **GPU Name:** ${L.deviceName}`),Z.push(`- **GPU Total Memory:** ${(L.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let F=Q.gpuTotalBytes>0?(Q.gpuUsableBytes/Q.gpuTotalBytes*100).toFixed(0):0;Z.push(`- **GPU Usable Memory:** ${(Q.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${F}% of ${(Q.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`)}}else Z.push("- **GPU:** Not available")}}Z.push(`
+### Command Used`);let w=process.argv.slice(2).join(" ");if(Z.push(`\`\`\`bash
+${process.argv[0]} ${process.argv[1]} ${w}
+\`\`\``),Z.push(`
+### Package Information`),Z.push(`- **Name:** ${"@fugood/buttress-server"}`),Z.push(`- **Version:** ${"2.23.0-beta.53"}`),X&&Object.keys(X).length>0){Z.push(`
+### Configuration`),Z.push("<details>"),Z.push("<summary>Click to expand TOML configuration</summary>"),Z.push("\n```toml");try{let U=P3.stringify(X);Z.push(U)}catch(U){Z.push("# Error serializing config"),Z.push(JSON.stringify(X,null,2))}Z.push("```"),Z.push("</details>")}let A=`ggml-stt-model-capabilities-${new Date().toISOString().replace(/[.:]/g,"-").split("T")[0]}.md`,R=S3.join(process.cwd(),A);x3.writeFileSync(R,Z.join(`
+`),"utf8"),console.log(`
+STT model capabilities table saved to: ${R}`),process.exit(0)}catch(W){console.error("Failed to generate STT model table:",W.message),process.exit(1)}}async function D3({modelId:$=null,defaultConfig:X=null}={}){if(console.log(`${"@fugood/buttress-server"} v${"2.23.0-beta.53"}`),console.log("Testing capabilities for backend: ggml-stt"),$)console.log(`Model: ${$}`);console.log("--------------------------------");try{let{repoId:Z,filename:W}=T3($),j=X||{},{server:V,generators:Y=[],...J}=j,N=(z={},A={})=>{let R=Array.isArray(z)?[...z]:{...z};return Object.entries(A||{}).forEach(([U,Q])=>{if(Q&&typeof Q==="object"&&!Array.isArray(Q))R[U]=N(R[U]||{},Q);else R[U]=Q}),R},H=(z)=>N(JSON.parse(JSON.stringify(J)),z||{}),q=((z)=>{if(Array.isArray(Y)&&Y.length>0){let A=Y.filter((R)=>R?.type==="ggml-stt");if(A.length>0){if(z){let R=A.find((U)=>U.model?.repo_id===z);if(R)return H(R)}}}if(Object.keys(J).length>0)return H({});return null})(Z);if(Z)q={...q||{},model:{...J.runtime,...q?.model||{},repo_id:Z,...W&&{filename:W}}};let _=await t("ggml-stt",null,{config:q,includeBreakdown:!0}),G=_.buttress?.selected||null,w=_.modelConfig||null;if(Z||w?.repoId){console.log(`
+=== Model Information ===`);let z=Z||w?.repoId;if(console.log(`Repository ID: ${z}`),W)console.log(`Filename: ${W}`);if(G?.modelBytes)console.log(`Model Size: ${(G.modelBytes/1024/1024).toFixed(1)} MB`);let A=G?.processingBytes||G?.kvCacheBytes;if(A)console.log(`Processing Buffer: ${(A/1024/1024).toFixed(1)} MB`);if(G?.modelBytes&&A)console.log(`Total Required Memory: ${((G.modelBytes+A)/1024/1024).toFixed(1)} MB`);else if(_.buttress?.selected?.fit){let{totalRequiredBytes:R}=_.buttress.selected.fit;console.log(`Total Required Memory: ${(R/1024/1024).toFixed(1)} MB`)}}if(_.buttress?.selected){let{selected:z}=_.buttress;console.log(`
+=== Hardware Information ===`);let A=null;if(process.platform!=="win32")try{A=K4("uname -a",{encoding:"utf8"}).trim()}catch{}if(A)console.log(`System: ${A}`);else console.log(`Hostname: ${c.hostname()}`),console.log(`OS: ${c.type()} ${c.release()}`);console.log(`Platform: ${z.platform}`),console.log(`CPU Cores: ${c.cpus().length}`),console.log(`Total System Memory: ${(c.totalmem()/1024/1024/1024).toFixed(2)} GB`);let R=z.cpuTotalBytes>0?(z.cpuUsableBytes/z.cpuTotalBytes*100).toFixed(0):0;if(console.log(`Usable CPU Memory: ${(z.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${R}% of ${(z.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),z.hasGpu)console.log(`
+--- GPU Details ---`),z.devices.filter((Q)=>Q.type==="gpu").forEach((Q)=>{console.log(`GPU Backend: ${Q.backend}`),console.log(`GPU Name: ${Q.deviceName}`),console.log(`GPU Total Memory: ${(Q.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);let K=z.gpuTotalBytes>0?(z.gpuUsableBytes/z.gpuTotalBytes*100).toFixed(0):0;if(console.log(`GPU Usable Memory: ${(z.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${K}% of ${(z.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),Q.metadata){if(Q.metadata.hasBFloat16)console.log("Supports BFloat16: Yes");if(Q.metadata.hasUnifiedMemory)console.log("Unified Memory: Yes")}});else console.log("GPU: Not available");if(console.log(`
+Backend Variant: ${z.variant}`),console.log(`Performance Score: ${z.score}`),z.fit)console.log(`
+--- Model Fit Analysis ---`),console.log(`Fits in GPU: ${z.fit.fitsInGpu?"Yes":"No"}`),console.log(`Fits in CPU: ${z.fit.fitsInCpu?"Yes":"No"}`),console.log(`Limiting Factor: ${z.fit.limiting}`)}console.log(`
+=== Full Capabilities JSON ===`),console.log(JSON.stringify(_,null,2)),process.exit(0)}catch(Z){console.error("Failed to get capabilities:",Z.message),process.exit(1)}}var T3=($)=>{if(!$)return{repoId:null,filename:null};let[X,Z]=$.split(":");return{repoId:X,filename:Z||null}};var b3=P(()=>{V4();M0()});var d9={};import y9 from"node:fs";import u9 from"node:path";import f9 from"@iarna/toml";async function g9($){if(!$?.generators||!Array.isArray($.generators))return;let X=$.generators.filter((O)=>{if(!O.model?.download)return!1;let{type:q}=O;if(!q||q!=="ggml-llm"&&q!=="ggml-stt")return console.warn(`[Download] Skipping unknown generator type: ${q}`),!1;return!0});if(X.length===0)return;let{server:Z,generators:W,...j}=$,V=X.map((O)=>{let{type:q}=O,_=O.model?.repo_id;console.log(`[Download] Starting pre-download for ${q}: ${_}`);let G={...j,backend:O.backend||{},model:O.model||{},runtime:{...j.runtime,...O.runtime||{}}};return Q4(q,G,{onProgress:()=>{},onComplete:({repoId:w,alreadyExists:z})=>{if(z)console.log(`[Download] Pre-download complete (already exists): ${w}`);else console.log(`[Download] Pre-download complete: ${w}`)},onError:(w)=>{console.error(`[Download] Pre-download failed for ${_}:`,w.message)}})}),Y=await Promise.all(V),J=Y.filter((O)=>O.started).length,N=Y.filter((O)=>O.alreadyExists).length,H=Y.filter((O)=>O.alreadyDownloading).length;console.log(`[Download] Pre-download summary: ${J} started, ${N} already exist, ${H} already downloading`)}var v3,h3,I3,q4,J0=null,p9,m9,C3,_4,c9;var y3=P(async()=>{b3();b0();q1();await g3();if(process.argv.includes("--version")||process.argv.includes("-v"))console.log("2.23.0-beta.53"),process.exit(0);if(process.argv.includes("--help")||process.argv.includes("-h"))console.log(`
+bricks-buttress v${"2.23.0-beta.53"}
+
+Buttress server for remote inference with GGML backends.
+
+Usage:
+  bricks-buttress [options]
+
+Options:
+  -h, --help                    Show this help message
+  -v, --version                 Show version number
+  -p, --port <port>             Port to listen on (default: 2080)
+  -c, --config <path|toml>      Path to TOML config file or inline TOML string
+
+Testing Options:
+  --test-caps <backend>         Test model capabilities (ggml-llm or ggml-stt)
+  --test-caps-model-id <id>     Model ID to test (used with --test-caps)
+  --test-models <ids>           Comma-separated list of model IDs to test
+  --test-models-default         Test default set of models
+
+  Note: --test-models and --test-models-default output a markdown report
+        file (e.g., ggml-llm-model-capabilities-YYYY-MM-DD.md)
+
+Environment Variables:
+  NODE_ENV                      Set to 'development' for dev mode
+  ENABLE_OPENAI_COMPAT_ENDPOINT Set to '1' to enable OpenAI-compatible API
+
+Examples:
+  bricks-buttress
+  bricks-buttress --port 3000
+  bricks-buttress --config ./config.toml
+  bricks-buttress --test-caps ggml-llm --test-models-default
+  bricks-buttress --test-caps ggml-stt --test-caps-model-id BricksDisplay/whisper-ggml:ggml-small.bin
+`),process.exit(0);v3=process.argv.findIndex(($)=>$==="--port"||$==="-p"),h3=v3>=0?Number(process.argv[v3+1]):void 0,I3=process.argv.findIndex(($)=>$==="--config"||$==="-c"),q4=I3>=0?process.argv[I3+1]:null;if(q4){let $;if(q4.includes(`
+`))$=q4;else{let X=u9.resolve(q4);try{$=y9.readFileSync(X,"utf8")}catch(Z){console.error(`Failed to read Buttress config at ${X}:`,Z),process.exit(1)}}try{let X=f9.parse($);if(X.env&&typeof X.env==="object")Object.entries(X.env).forEach(([Z,W])=>{if(process.env[Z]===void 0)process.env[Z]=String(W)}),delete X.env;J0=X}catch(X){console.error("Failed to parse TOML config:",X),process.exit(1)}}p9=["ggml-org/gpt-oss-20b-GGUF","ggml-org/gpt-oss-120b-GGUF","unsloth/Nemotron-3-Nano-30B-A3B-GGUF","unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF","unsloth/GLM-4.7-Flash-GGUF","bartowski/Mistral-Nemo-Instruct-2407-GGUF","mistralai/Magistral-Small-2509-GGUF","mistralai/Ministral-3-14B-Reasoning-2512-GGUF","bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF","bartowski/mistralai_Devstral-2-123B-Instruct-2512-GGUF","ggml-org/gemma-3-12b-it-qat-GGUF","ggml-org/gemma-3-27b-it-qat-GGUF","unsloth/phi-4-GGUF"],m9=["BricksDisplay/whisper-ggml:ggml-small.bin","BricksDisplay/whisper-ggml:ggml-small-q8_0.bin","BricksDisplay/whisper-ggml:ggml-medium.bin","BricksDisplay/whisper-ggml:ggml-medium-q8_0.bin","BricksDisplay/whisper-ggml:ggml-large-v3-turbo.bin","BricksDisplay/whisper-ggml:ggml-large-v3-turbo-q8_0.bin","BricksDisplay/whisper-ggml:ggml-large-v3.bin"],C3=process.argv.findIndex(($)=>$==="--test-caps");if(C3>=0){let $=process.argv[C3+1]||"ggml-llm";if($!=="ggml-llm"&&$!=="ggml-stt")console.error("Only ggml-llm and ggml-stt backends are supported for testing capabilities"),process.exit(1);let X=process.argv.findIndex((W)=>W==="--test-models"),Z=process.argv.includes("--test-models-default");if($==="ggml-stt")if(X>=0){let W=process.argv[X+1];if(!W)console.error("Error: --test-models requires a comma-separated list of model IDs"),process.exit(1);let j=W.split(",").map((V)=>V.trim());await L1({modelIds:j,defaultConfig:J0})}else if(Z)await L1({modelIds:m9,defaultConfig:J0});else{let W=process.argv.findIndex((V)=>V==="--test-caps-model-id"),j=W>=0?process.argv[W+1]:null;await D3({modelId:j,defaultConfig:J0})}else if(X>=0){let W=process.argv[X+1];if(!W)console.error("Error: --test-models requires a comma-separated list of model IDs"),process.exit(1);let j=W.split(",").map((V)=>V.trim());await _1({modelIds:j,defaultConfig:J0})}else if(Z)await _1({modelIds:p9,defaultConfig:J0});else{let W=process.argv.findIndex((V)=>V==="--test-caps-model-id"),j=W>=0?process.argv[W+1]:null;await k3({modelId:j,defaultConfig:J0})}}_4=G1(J0);if(h3)_4.server.port=h3;if(!_4.server.port)_4.server.port=2080;c9=process.env.ENABLE_OPENAI_COMPAT_ENDPOINT==="1";f3({config:_4,enableOpenAICompat:c9}).then(async({port:$,openaiEnabled:X,autoDiscover:Z})=>{let W=O4();if(console.log(`Buttress server listening on port ${$}`),console.log("--------------------------------"),await u3(),console.log(),console.log("Current supported Generators:"),console.log("- LLM (GGML)"),console.log("- STT (GGML)"),console.log(),console.log("Please configure `Buttress (Remote Inference)` in the Generator to connect to this server."),console.log(),console.log(`- Use http://${W}:${$} to connect to this server via LAN.`),console.log(`- Visit http://${W}:${$}/status to see status via LAN.`),console.log(),X)console.log("OpenAI-compatible API [EXPERIMENTAL]:"),console.log(`- Base URL: http://${W}:${$}/oai-compat/v1`),console.log(`- Chat completions: POST http://${W}:${$}/oai-compat/v1/chat/completions`),console.log(`- Models: GET http://${W}:${$}/oai-compat/v1/models`),console.log();else console.log("OpenAI-compatible API [EXPERIMENTAL]: disabled"),console.log("  Set ENABLE_OPENAI_COMPAT_ENDPOINT=1 to enable"),console.log();if(Z)console.log("Auto-discover enabled"),console.log();if(J0)await g9(J0)}).catch(($)=>{console.error("Failed to start Buttress server:",$),process.exitCode=1})});import{node as l9}from"@elysiajs/node";import{Elysia as i9,t as v0}from"elysia";import{ReadableStream as s9}from"node:stream/web";import n9 from"node:fs/promises";import{ZodError as r9}from"zod";var o9=async()=>{let $=`https://registry.npmjs.org/${"@fugood/buttress-server"}/latest`;try{let X=new AbortController,Z=setTimeout(()=>X.abort(),3000),W=await fetch($,{headers:{Accept:"application/json"},signal:X.signal});if(clearTimeout(Z),!W.ok)return null;return(await W.json()).version||null}catch{return null}},a9=($,X)=>{if(!X)return!1;let Z=$.split(/[.-]/),W=X.split(/[.-]/);for(let j=0;j<Math.max(Z.length,W.length);j+=1){let V=parseInt(Z[j])||0,Y=parseInt(W[j])||0;if(Y>V)return!0;if(Y<V)return!1}return!1},t9=($)=>{console.log(""),console.log("\x1B[33m╭─────────────────────────────────────────────────╮\x1B[0m"),console.log("\x1B[33m│\x1B[0m  Update available! \x1B[2m%s\x1B[0m → \x1B[32m%s\x1B[0m","2.23.0-beta.53".padEnd(12),$.padEnd(12),"\x1B[33m│\x1B[0m"),console.log("\x1B[33m│\x1B[0m                                                 \x1B[33m│\x1B[0m"),console.log("\x1B[33m│\x1B[0m  Run to upgrade:                                \x1B[33m│\x1B[0m"),console.log("\x1B[33m│\x1B[0m  \x1B[36mnpm install -g %s\x1B[0m      \x1B[33m│\x1B[0m","@fugood/buttress-server".padEnd(27)),console.log("\x1B[33m╰─────────────────────────────────────────────────╯\x1B[0m"),console.log("")},u3=async()=>{try{let $=await o9();if($&&a9("2.23.0-beta.53",$))t9($)}catch($){}},e9,$$=async({backend:$,router:X,config:Z,enableOpenAICompat:W})=>{try{await n9.mkdir(Z.server.temp_file_dir,{recursive:!0})}catch{}let j=O4()||"0.0.0.0",V={id:Z.server.id,name:Z.server.name,version:"2.23.0-beta.53",address:j,port:Z.server.port,url:`http://${j}:${Z.server.port}`,generators:V3(Z,Z.generators.map((N)=>N.type)),authentication:{required:!0,type:"device-group"}},Y=new i9({serve:{maxRequestBodySize:Z.server.max_body_size},websocket:{idleTimeout:Math.ceil(Z.server.session_timeout/1000)},adapter:e9?l9():void 0}).state({sessions:new Map,backend:$||J4,config:Z,serverInfo:V});if(X)Y.use(X);if(Z.autodiscover?.http?.enabled)Y.use(j1(Z));if(Y.use(Q1),Y.use(V1),W)Y.use(U4(Z));let J={INVALID_REQUEST:-32600,INVALID_PARAMS:-32602,METHOD_NOT_FOUND:-32601,INTERNAL_ERROR:-32603};return Y.ws("/buttress/rpc",{parse:(N,H)=>{if(typeof H==="string")try{return JSON.parse(H)}catch{return N.send(JSON.stringify({jsonrpc:"2.0",error:{code:J.INVALID_REQUEST,message:"Invalid request"},id:null})),null}return H},body:v0.Object({jsonrpc:v0.String(),method:v0.String(),params:v0.String(),id:v0.String()}),open(N){let H=N.id??N.raw?.id??N.remoteAddress;if(console.log(`[Request] New connection: ${H}`),!N.data.store.sessions.has(H))N.data.store.sessions.set(H,{streamReaders:new Map,generators:new Set,initializedContexts:new Set,timeout:null});else{let O=N.data.store.sessions.get(H);clearTimeout(O.timeout),O.timeout=null}},async message(N,{id:H,method:O,params:q}){let _=N.id??N.raw?.id??N.remoteAddress;console.log(`[Request] Received request from ${_}: ${O}`);let G=N.data.store.sessions.get(_),[w,z]=O.split("."),A=z3[w]?.[z];if(!A){N.send(JSON.stringify({jsonrpc:"2.0",error:{code:J.METHOD_NOT_FOUND,message:"Method not found"},id:H}));return}try{if(O==="cancel"){if(G.streamReaders.has(H))G.streamReaders.get(H)?.cancel(),G.streamReaders.delete(H);return}if(O==="ping"){N.send(JSON.stringify({jsonrpc:"2.0",result:"pong",id:H}));return}let R=R3(q),U=A3[w]?.[z],Q=U?U.parse(R):R,K={...N.data.store,peerId:_,session:G},B=await A(K,...Q);if(B instanceof s9){let L=B.getReader();G.streamReaders.set(H,L),N.send(JSON.stringify({jsonrpc:"2.0",result:{type:"stream"},id:H}));try{while(!0){let{value:F,done:E}=await L.read();if(E)break;let{event:M,data:x}=F;N.send(JSON.stringify({jsonrpc:"2.0",method:`notification/${M}`,params:H4(x),id:H}))}N.send(JSON.stringify({jsonrpc:"2.0",method:"notification/_end",id:H}))}catch(F){console.error(F),N.send(JSON.stringify({jsonrpc:"2.0",method:"notification/_error",params:H4(F),id:H}))}G.streamReaders.delete(H)}else N.send(JSON.stringify({jsonrpc:"2.0",result:H4(B),id:H}))}catch(R){if(R instanceof r9){N.send(JSON.stringify({jsonrpc:"2.0",error:{code:J.INVALID_PARAMS,message:"Invalid params",data:R.issues},id:H}));return}console.error(R),N.send(JSON.stringify({jsonrpc:"2.0",error:{code:J.INTERNAL_ERROR,message:String(R)},id:H}))}},async close(N){let H=N.id??N.raw?.id??N.remoteAddress;console.log(`[Request] Connection closed: ${H}`);let{backend:O,sessions:q}=N.data.store,_=q.get(H);if(!_)return;_.streamReaders.forEach((G)=>G.cancel()),_.streamReaders.clear(),_.timeout=setTimeout(()=>{q.delete(H),console.log(`[Request] Session timed out: ${H}`);let{generators:G}=_;G.forEach((w)=>{O.finalizeGenerator(w)})},Z.server.session_timeout)}}),{app:Y,config:Z}},f3=async({backend:$,router:X,config:Z,enableOpenAICompat:W=!1})=>{let{app:j,config:V}=await $$({backend:$,router:X,config:Z,enableOpenAICompat:W}),{server:{port:Y}}=V,J=[new Promise((H)=>j.listen(Y,H))],N=null;if(V.autodiscover)N=new K1(V.autodiscover,()=>j.store.serverInfo),J.push(N.start());return await Promise.all(J),{app:j,port:Y,openaiEnabled:W,autoDiscover:N}};var g3=P(async()=>{V4();Z3();B3();w3();b0();M3();q1();V4();b0();e9=typeof process<"u"&&process.versions&&process.versions.node;if(R1.main==R1.module)await y3().then(() => d9)});await g3();export{f3 as startServer,Q4 as startModelDownload,G1 as processConfig,t9 as logUpdateMessage,$$ as createServer,a9 as compareVersions,o9 as checkForUpdates,u3 as checkAndNotifyUpdates};