@lunora/ai 1.0.0-alpha.26 → 1.0.0-alpha.28

package/dist/index.d.mts

@@ -1,5 +1,5 @@
-import { L as LunoraAiOptions, a as LunoraAi } from "./packem_shared/types.d-F_Q8R8L3.mjs";
-export { A as AI_GATEWAY_ACCOUNT_ID_ENV, b as AI_GATEWAY_ID_ENV, c as AI_GATEWAY_TOKEN_ENV, type d as AiBindingLike, type e as AiGatewayMetadata, type f as AiGatewayOptions, type E as EmbeddingModelInput, type M as ModelInput, type R as ResolvedAiGateway, type W as WorkersAiProviderLike, g as buildAiGatewayMetadataFields, r as resolveAiGateway } from "./packem_shared/types.d-F_Q8R8L3.mjs";
+import { L as LunoraAiOptions, a as LunoraAi } from "./packem_shared/types.d-BcLGTChd.mjs";
+export { A as AI_GATEWAY_ACCOUNT_ID_ENV, b as AI_GATEWAY_ID_ENV, c as AI_GATEWAY_TOKEN_ENV, type d as AiBindingLike, type e as AiGatewayMetadata, type f as AiGatewayOptions, type E as EmbeddingModelInput, type M as ModelInput, type R as ResolvedAiGateway, type W as WorkersAiProviderLike, g as buildAiGatewayMetadataFields, r as resolveAiGateway } from "./packem_shared/types.d-BcLGTChd.mjs";
 export { type EmbeddingModel, type LanguageModel, embed, embedMany, generateObject, generateText, hasToolCall, jsonSchema, streamObject, streamText, tool } from 'ai';
 export { createWorkersAI } from 'workers-ai-provider';
 /**

package/dist/index.d.ts

@@ -1,5 +1,5 @@
-import { L as LunoraAiOptions, a as LunoraAi } from "./packem_shared/types.d-F_Q8R8L3.js";
-export { A as AI_GATEWAY_ACCOUNT_ID_ENV, b as AI_GATEWAY_ID_ENV, c as AI_GATEWAY_TOKEN_ENV, type d as AiBindingLike, type e as AiGatewayMetadata, type f as AiGatewayOptions, type E as EmbeddingModelInput, type M as ModelInput, type R as ResolvedAiGateway, type W as WorkersAiProviderLike, g as buildAiGatewayMetadataFields, r as resolveAiGateway } from "./packem_shared/types.d-F_Q8R8L3.js";
+import { L as LunoraAiOptions, a as LunoraAi } from "./packem_shared/types.d-BcLGTChd.js";
+export { A as AI_GATEWAY_ACCOUNT_ID_ENV, b as AI_GATEWAY_ID_ENV, c as AI_GATEWAY_TOKEN_ENV, type d as AiBindingLike, type e as AiGatewayMetadata, type f as AiGatewayOptions, type E as EmbeddingModelInput, type M as ModelInput, type R as ResolvedAiGateway, type W as WorkersAiProviderLike, g as buildAiGatewayMetadataFields, r as resolveAiGateway } from "./packem_shared/types.d-BcLGTChd.js";
 export { type EmbeddingModel, type LanguageModel, embed, embedMany, generateObject, generateText, hasToolCall, jsonSchema, streamObject, streamText, tool } from 'ai';
 export { createWorkersAI } from 'workers-ai-provider';
 /**

package/dist/index.mjs

@@ -1 +1 @@
-import{default as a}from"./packem_shared/createAi-k_AcrXgH.mjs";import{AI_GATEWAY_ACCOUNT_ID_ENV as o,AI_GATEWAY_ID_ENV as A,AI_GATEWAY_TOKEN_ENV as _,buildAiGatewayMetadataFields as m,resolveAiGateway as l}from"./packem_shared/AI_GATEWAY_ACCOUNT_ID_ENV-JzYYnjLP.mjs";import{embed as T,embedMany as E,generateObject as d,generateText as x,hasToolCall as I,jsonSchema as b,streamObject as c,streamText as f,tool as i}from"ai";import{createWorkersAI as N}from"workers-ai-provider";export{o as AI_GATEWAY_ACCOUNT_ID_ENV,A as AI_GATEWAY_ID_ENV,_ as AI_GATEWAY_TOKEN_ENV,m as buildAiGatewayMetadataFields,a as createAi,N as createWorkersAI,T as embed,E as embedMany,d as generateObject,x as generateText,I as hasToolCall,b as jsonSchema,l as resolveAiGateway,c as streamObject,f as streamText,i as tool};
+import{default as a}from"./packem_shared/createAi-C2ExoUDR.mjs";import{AI_GATEWAY_ACCOUNT_ID_ENV as o,AI_GATEWAY_ID_ENV as A,AI_GATEWAY_TOKEN_ENV as _,buildAiGatewayMetadataFields as m,resolveAiGateway as l}from"./packem_shared/AI_GATEWAY_ACCOUNT_ID_ENV-CPzc-So1.mjs";import{embed as T,embedMany as E,generateObject as d,generateText as x,hasToolCall as I,jsonSchema as b,streamObject as c,streamText as f,tool as i}from"ai";import{createWorkersAI as N}from"workers-ai-provider";export{o as AI_GATEWAY_ACCOUNT_ID_ENV,A as AI_GATEWAY_ID_ENV,_ as AI_GATEWAY_TOKEN_ENV,m as buildAiGatewayMetadataFields,a as createAi,N as createWorkersAI,T as embed,E as embedMany,d as generateObject,x as generateText,I as hasToolCall,b as jsonSchema,l as resolveAiGateway,c as streamObject,f as streamText,i as tool};

package/dist/packem_shared/AI_GATEWAY_ACCOUNT_ID_ENV-CPzc-So1.mjs

@@ -0,0 +1 @@
+const r=(t,a)=>{const n=t[a];return typeof n=="string"&&n.length>0?n:void 0};let c=!1;const u=t=>{if(t===void 0)return;const a={};return typeof t.functionPath=="string"&&t.functionPath.length>0&&(a.functionPath=t.functionPath),typeof t.traceId=="string"&&t.traceId.length>0&&(a.traceId=t.traceId),Object.keys(a).length>0?a:void 0},h=t=>{const a=u(t);return a===void 0?void 0:JSON.stringify(a)},I="LUNORA_AI_GATEWAY_ACCOUNT_ID",_="LUNORA_AI_GATEWAY_ID",A="LUNORA_AI_GATEWAY_TOKEN",f=(t,a,n="byo-provider")=>{const o=r(t,I),e=r(t,_);if(o===void 0||e===void 0)return;const s=r(t,A),i={};s!==void 0&&(i["cf-aig-authorization"]=`Bearer ${s}`,n==="workers-ai-binding"&&!c&&(c=!0,console.warn(`[lunora:ai] ${A} is set, but the Workers AI binding cannot send a gateway auth token — Cloudflare's native gateway option has no authorization field. The token is ignored on this path; use a bring-your-own AI SDK provider (which sends cf-aig-authorization), or make the AI Gateway unauthenticated for Workers AI.`)));const d=h(a);return d!==void 0&&(i["cf-aig-metadata"]=d),{accountId:o,baseURL:`https://gateway.ai.cloudflare.com/v1/${o}/${e}`,gatewayId:e,headers:i}};export{I as AI_GATEWAY_ACCOUNT_ID_ENV,_ as AI_GATEWAY_ID_ENV,A as AI_GATEWAY_TOKEN_ENV,u as buildAiGatewayMetadataFields,f as resolveAiGateway};

package/dist/packem_shared/createAi-C2ExoUDR.mjs

@@ -0,0 +1 @@
+import{LunoraError as t}from"@lunora/errors";import{createWorkersAI as c}from"workers-ai-provider";import{buildAiGatewayMetadataFields as v,resolveAiGateway as A}from"./AI_GATEWAY_ACCOUNT_ID_ENV-CPzc-So1.mjs";const I=(i,o,s)=>{const r=v(s);if(i!==void 0)return r!==void 0&&i.metadata===void 0?{...i,metadata:r}:i;if(o===void 0)return;const a=A(o,s,"workers-ai-binding");if(a!==void 0)return r===void 0?{id:a.gatewayId}:{id:a.gatewayId,metadata:r}},y=(i,o)=>c({binding:i,gateway:o}),h=i=>{const{binding:o,defaultEmbeddingModel:s,defaultModel:r,env:a,gateway:f,metadata:g,provider:m}=i;if(!m&&!o)throw new t("INTERNAL","@lunora/ai: createAi requires a `binding` (env.AI) or a pre-built `provider`");const u=I(f,a,g),n=m??y(o,u),p=e=>{if(e===void 0){if(!r)throw new t("INTERNAL","@lunora/ai: no model supplied and no `defaultModel` configured — pass a model id or an AI SDK model");return n(r)}return typeof e=="string"?n(e):e},w=e=>{const d=n.textEmbeddingModel;if(typeof d!="function")throw new t("INTERNAL","@lunora/ai: the Workers AI provider does not expose `textEmbeddingModel`; pass an AI SDK EmbeddingModel (e.g. from @ai-sdk/openai) to embed()");return d.call(n,e)};return{embeddingModel:e=>{if(typeof e=="object")return e;const d=e??s;if(!d)throw new t("INTERNAL","@lunora/ai: no embedding model supplied and no `defaultEmbeddingModel` configured — pass an embedding model id or an AI SDK EmbeddingModel");return w(d)},model:p,run:async(e,d,l)=>{if(!o)throw new t("INTERNAL","@lunora/ai: ai.run requires the `binding` (env.AI) — it is unavailable when only a custom `provider` was supplied");const b=u!==void 0&&l?.gateway===void 0?{...l,gateway:u}:l;return o.run(e,d,b)},workersai:n}};export{h as default};

package/dist/packem_shared/{defineRag-BF3RwHZj.mjs → defineRag-DUOcnSgC.mjs}

@@ -1,8 +1,8 @@
-import{LunoraError as x}from"@lunora/errors";import{tool as G,jsonSchema as J,embed as ee}from"ai";import te from"./fixedWindowChunks-XJRXHEoz.mjs";import ne from"./hybridRank-U6PmGuz1.mjs";const ae=8,re=async(e,s,h)=>{if(!Number.isInteger(s)||s<1)throw new RangeError("concurrentMap: `limit` must be a positive integer");if(e.length===0)return[];const b=Math.max(1,Math.min(s,e.length)),y=Array.from({length:e.length});let f=0,k=!1,v;const E=async()=>{for(;;){if(k)return;const I=f;if(f+=1,I>=e.length)return;try{y[I]=await h(e[I],I)}catch(N){k||(k=!0,v=N);return}}},m=Array.from({length:b},()=>E());if(await Promise.all(m),k)throw v;return y},oe=1e3,ie=200,se=5,ce=20,de=100,z="__ragChunk",V="__ragSource",A="__ragText",R="__ragHash",T="__ragChunks",M="__ragImportance",F="__ragModel",ue=new Set([z,T,R,M,F,V,A]),le=/^[\w.-]{1,40}$/,P=e=>e===void 0?"":`${encodeURIComponent(e)}#`,_=(e,s,h)=>`${P(e)}${s}#${String(h)}`,K=(e,s)=>{const h=P(s),b=h!==""&&e.startsWith(h)?e.slice(h.length):e,y=b.lastIndexOf("#"),f=y===-1?Number.NaN:Number(b.slice(y+1));return y===-1||!Number.isInteger(f)||f<0?{chunkIndex:0,sourceId:b}:{chunkIndex:f,sourceId:b.slice(0,y)}},me=async e=>{const s=await crypto.subtle.digest("SHA-256",new TextEncoder().encode(e));return[...new Uint8Array(s)].map(h=>h.toString(16).padStart(2,"0")).join("")},O=e=>{if(!e)return;const s=Object.entries(e).filter(([h])=>!ue.has(h));return s.length>0?Object.fromEntries(s):void 0},Q=new Set,he=e=>{Q.has(e)||(Q.add(e),console.warn(`[@lunora/ai/rag] index "${e}" is used without a namespace — in a multi-tenant/sharded
+import{LunoraError as x}from"@lunora/errors";import{tool as G,jsonSchema as J,embed as ee}from"ai";import te from"./fixedWindowChunks-XJRXHEoz.mjs";import ne from"./hybridRank-U6PmGuz1.mjs";const ae=8,re=async(e,s,p)=>{if(!Number.isInteger(s)||s<1)throw new RangeError("concurrentMap: `limit` must be a positive integer");if(e.length===0)return[];const b=Math.max(1,Math.min(s,e.length)),y=Array.from({length:e.length});let f=0,k=!1,v;const E=async()=>{for(;;){if(k)return;const I=f;if(f+=1,I>=e.length)return;try{y[I]=await p(e[I],I)}catch(N){k||(k=!0,v=N);return}}},h=Array.from({length:b},()=>E());if(await Promise.all(h),k)throw v;return y},oe=1e3,ie=200,se=5,ce=20,de=100,z="__ragChunk",V="__ragSource",A="__ragText",R="__ragHash",T="__ragChunks",M="__ragImportance",F="__ragModel",ue=new Set([z,T,R,M,F,V,A]),me=/^[\w.-]{1,40}$/,P=e=>e===void 0?"":`${encodeURIComponent(e)}#`,_=(e,s,p)=>`${P(e)}${s}#${String(p)}`,K=(e,s)=>{const p=P(s),b=p!==""&&e.startsWith(p)?e.slice(p.length):e,y=b.lastIndexOf("#"),f=y===-1?Number.NaN:Number(b.slice(y+1));return y===-1||!Number.isInteger(f)||f<0?{chunkIndex:0,sourceId:b}:{chunkIndex:f,sourceId:b.slice(0,y)}},he=async e=>{const s=await crypto.subtle.digest("SHA-256",new TextEncoder().encode(e));return[...new Uint8Array(s)].map(p=>p.toString(16).padStart(2,"0")).join("")},O=e=>{if(!e)return;const s=Object.entries(e).filter(([p])=>!ue.has(p));return s.length>0?Object.fromEntries(s):void 0},Q=new Set,pe=e=>{Q.has(e)||(Q.add(e),console.warn(`[@lunora/ai/rag] index "${e}" is used without a namespace — in a multi-tenant/sharded
 app this shares one tenant's chunks (text included) with every other tenant, since
 Vectorize indexes are account-global. Pass \`namespace\` (the shard/tenant key) on both
-index() and retrieve(). Single-tenant apps suppress this via { allowSharedNamespace: true }.`))},pe=e=>e.map(s=>`[source:${s.sourceId}#${String(s.chunkIndex)}]
+index() and retrieve(). Single-tenant apps suppress this via { allowSharedNamespace: true }.`))},le=e=>e.map(s=>`[source:${s.sourceId}#${String(s.chunkIndex)}]
 ${s.text}`).join(`
 
-`),ge=(e,s)=>{if(typeof e=="object")return e;if(s===void 0)throw new x("INTERNAL","@lunora/ai/rag: `embeddingModel` is a Workers AI model id (or omitted) but the bound context has no `ai` (env.AI). Pass an AI SDK EmbeddingModel object to embed without Workers AI, or bind a context whose `ctx.ai` is wired.");return s.embeddingModel(e)},fe=e=>{const s=e.modelId;return typeof s=="string"&&s.length>0?s:void 0},we=e=>{if(!(typeof e!="object"||e===null)){for(const s of Object.values(e))if(typeof s=="object"&&s!==null){const{cost:h}=s;if(typeof h=="number"&&Number.isFinite(h))return h}}},Ie=e=>{if(typeof e.index!="string"||e.index.length===0)throw new x("BAD_REQUEST","@lunora/ai/rag: `index` must be a non-empty Vectorize index name");const s=e.chunkSize??oe,h=e.chunkOverlap??ie;if(!Number.isInteger(s)||s<1)throw new x("BAD_REQUEST","@lunora/ai/rag: `chunkSize` must be a positive integer");if(!Number.isInteger(h)||h<0||h>=s)throw new x("BAD_REQUEST","@lunora/ai/rag: `chunkOverlap` must be a non-negative integer smaller than `chunkSize`");const b=e.topK??se;if(!Number.isInteger(b)||b<1)throw new x("BAD_REQUEST","@lunora/ai/rag: `topK` must be a positive integer");if(e.embeddingModelVersion!==void 0&&!le.test(e.embeddingModelVersion))throw new x("BAD_REQUEST",'@lunora/ai/rag: `embeddingModelVersion` must match /^[A-Za-z0-9._-]{1,40}$/ (a short, stable tag like "bge-v1.5")');const y=e.chunk??(m=>te(m,s,h)),{textStore:f}=e,k=f?de:ce,v=e.embeddingModelVersion,E=m=>v===void 0?m:m===void 0?v:`${v}::${m}`;return m=>{let I;const N=typeof m.trace=="function"?m.trace:void 0,B=async t=>{I??=ge(e.embeddingModel,m.ai);const n=I,o=async d=>{const{embedding:r,providerMetadata:u,usage:l}=await ee({model:n,value:t});if(d!==void 0){const c=l.tokens;typeof c=="number"&&Number.isFinite(c)&&d.setAttribute("gen_ai.usage.input_tokens",c);const p=we(u);p!==void 0&&d.setAttribute("gen_ai.usage.cost",p)}return r};if(N===void 0)return o();const i=fe(n),a=typeof m.conversationId=="string"&&m.conversationId.length>0?m.conversationId:void 0;return N("ai.embed",(d,r)=>o(r),{"gen_ai.operation.name":"embeddings",...i===void 0?{}:{"gen_ai.request.model":i},...a===void 0?{}:{"gen_ai.conversation.id":a}})},$=t=>{if(t===void 0){if(e.requireNamespace)throw new x("BAD_REQUEST",`@lunora/ai/rag: index "${e.index}" requires a namespace (requireNamespace is set) — pass the tenant/shard key on index()/retrieve()/remove()`);e.allowSharedNamespace||he(e.index)}},U=async(t,n)=>{const[o]=await m.vectors.getByIds(e.index,[_(n,t,0)],n),i=o?.metadata?.[R],a=o?.metadata?.[T];return{chunks:typeof a=="number"&&Number.isInteger(a)&&a>0?a:void 0,hash:typeof i=="string"?i:void 0}},j=async(t,n,o,i)=>{const a=Array.from({length:o-n},(d,r)=>_(i,t,n+r));a.length!==0&&(await m.vectors.deleteByIds(e.index,a,i),await f?.remove?.(a,{namespace:i}),await e.lexicalStore?.remove?.(a,{namespace:i}))},W=async t=>{if($(t.namespace),t.importance!==void 0&&(typeof t.importance!="number"||t.importance<0||t.importance>1))throw new x("BAD_REQUEST","@lunora/ai/rag: `importance` must be a number in [0, 1]");const n=E(t.namespace),o=await me(t.text),i=await U(t.id,n);if(i.hash===o&&i.chunks!==void 0)return{chunks:i.chunks,ids:Array.from({length:i.chunks},(r,u)=>_(n,t.id,u)),unchanged:!0};const a=y(t.text),d=a.map((r,u)=>_(n,t.id,u));if(a.length===0&&t.allowEmptySources===!1)throw new x("BAD_REQUEST",`@lunora/ai/rag: source "${t.id}" produced zero chunks — set allowEmptySources: true to allow this`);if(a.length>0){const r=a.map((u,l)=>({chunkIndex:l,id:d[l],sourceId:t.id,text:u}));f&&await f.put(r,{namespace:n}),e.lexicalStore&&await e.lexicalStore.index(r,{namespace:n})}return await re(a,ae,async(r,u)=>{const l=d[u],c={...t.metadata,[z]:u,[V]:t.id};f||(c[A]=r),t.importance!==void 0&&(c[M]=t.importance),u===0&&(c[R]=o,c[T]=a.length,v!==void 0&&(c[F]=v)),await m.vectors.upsert(e.index,{embed:B,id:l,input:r,metadata:c,namespace:n}),t.onChunk?.({chunkIndex:u,id:l,text:r,total:a.length})}),i.chunks!==void 0&&i.chunks>a.length&&await j(t.id,a.length,i.chunks,n),{chunks:a.length,ids:d,unchanged:!1}},H=async t=>{$(t.namespace);const n=E(t.namespace),o=(await U(t.id,n)).chunks??1;await j(t.id,0,o,n)},q=async(t,n)=>{const o=new Map;if(t.length===0)return o;if(f){const a=await f.getMany(t,{namespace:n});for(const[d,r]of t.entries()){const u=a[d];typeof u=="string"&&o.set(r,u)}return o}const i=await m.vectors.getByIds(e.index,t,n);for(const a of i){const d=a.metadata?.[A];typeof d=="string"&&o.set(a.id,d)}return o},L=async(t,n,o)=>{const i=n?.chunkContext?.before??0,a=n?.chunkContext?.after??0;if(i===0&&a===0)return t;if(!Number.isInteger(i)||i<0||!Number.isInteger(a)||a<0)throw new x("BAD_REQUEST","@lunora/ai/rag: `chunkContext.before`/`chunkContext.after` must be non-negative integers");const d=new Map(t.map(c=>[c.id,c.text])),r=new Set;for(const c of t)for(let p=-i;p<=a;p+=1){const w=c.chunkIndex+p,g=_(o,c.sourceId,w);p!==0&&w>=0&&!d.has(g)&&r.add(g)}const u=await q([...r],o),l=(c,p)=>{const w=_(o,c,p);return d.get(w)??u.get(w)};return t.map(c=>{const p=[];for(let w=-i;w<=a;w+=1){const g=w===0?c.text:l(c.sourceId,c.chunkIndex+w);g!==void 0&&p.push(g)}return{...c,text:p.join(`
-`)}})},X=t=>{if(typeof t=="string"){const n=e.filters?.[t];if(!n)throw new x("NOT_FOUND",`@lunora/ai/rag: unknown named filter "${t}" — must be one of the keys declared in RagConfig.filters`);return n.filter}return t},Y=(t,n)=>t.matches.map(o=>{const i=o.metadata??{},a=K(o.id,n),d=i[A],r=i[M],u=typeof r=="number"&&r>=0&&r<=1?r:1;return{chunkIndex:a.chunkIndex,id:o.id,importance:u,metadata:O(i),score:o.score*u,sourceId:a.sourceId,text:typeof d=="string"?d:""}}),Z=async(t,n)=>{if(!f)return t;const o=t.map(r=>r.id),[i,a]=await Promise.all([q(o,n),m.vectors.getByIds(e.index,o,n)]),d=new Map(a.map(r=>[r.id,r.metadata]));return t.flatMap(r=>{const u=i.get(r.id);if(u===void 0)return[];const l=d.get(r.id),c=l?.[M],p=typeof c=="number"&&c>=0&&c<=1?c:r.importance,w=r.score/r.importance*p;return[{...r,importance:p,metadata:O(l)??r.metadata,score:w,text:u}]})},C=async(t,n)=>{$(n?.namespace);const o=E(n?.namespace),i=X(n?.filter),a=e.rlsFilter?await e.rlsFilter(m.auth):void 0,d=a?{...i,...a}:i,r=Math.min(n?.topK??b,k),u=await m.vectors.query(e.index,{embed:B,filter:d,input:t,namespace:o,returnMetadata:f?"indexed":"all",topK:r});let l=await Z(Y(u,o),o);const c=n?.minScore;if(c!==void 0&&(l=l.filter(g=>g.score>=c)),e.lexicalStore){const g=(await e.lexicalStore.search(t,{filter:d,namespace:o,topK:e.lexicalTopK??r})).map(S=>{const D=K(S.id,o);return{chunkIndex:D.chunkIndex,id:S.id,importance:1,metadata:void 0,score:S.score,sourceId:D.sourceId,text:S.text}});l=[...ne(l,g)]}l.sort((g,S)=>S.score-g.score),l=[...await L(l,n,o)];const p=[],w=new Set;for(const g of l)w.has(g.sourceId)||(w.add(g.sourceId),p.push({id:g.sourceId,metadata:g.metadata,weight:g.importance}));return n?.onRetrieve?.({matches:l.length,query:t}),{chunks:l,context:pe(l),sources:p}};return{asTool:t=>G({description:t?.description??`Search the "${e.index}" knowledge base for passages relevant to a natural-language query.`,execute:async({query:n})=>C(n,{namespace:t?.namespace,topK:t?.topK}),inputSchema:J({properties:{query:{description:"The natural-language search query.",type:"string"}},required:["query"],type:"object"})}),index:W,remove:H,retrieve:C}}};export{Ie as default};
+`),ge=(e,s)=>{if(typeof e=="object")return e;if(s===void 0)throw new x("INTERNAL","@lunora/ai/rag: `embeddingModel` is a Workers AI model id (or omitted) but the bound context has no `ai` (env.AI). Pass an AI SDK EmbeddingModel object to embed without Workers AI, or bind a context whose `ctx.ai` is wired.");return s.embeddingModel(e)},fe=e=>{const s=e.modelId;return typeof s=="string"&&s.length>0?s:void 0},we=e=>{if(!(typeof e!="object"||e===null)){for(const s of Object.values(e))if(typeof s=="object"&&s!==null){const{cost:p}=s;if(typeof p=="number"&&Number.isFinite(p))return p}}},Ie=e=>{if(typeof e.index!="string"||e.index.length===0)throw new x("BAD_REQUEST","@lunora/ai/rag: `index` must be a non-empty Vectorize index name");const s=e.chunkSize??oe,p=e.chunkOverlap??ie;if(!Number.isInteger(s)||s<1)throw new x("BAD_REQUEST","@lunora/ai/rag: `chunkSize` must be a positive integer");if(!Number.isInteger(p)||p<0||p>=s)throw new x("BAD_REQUEST","@lunora/ai/rag: `chunkOverlap` must be a non-negative integer smaller than `chunkSize`");const b=e.topK??se;if(!Number.isInteger(b)||b<1)throw new x("BAD_REQUEST","@lunora/ai/rag: `topK` must be a positive integer");if(e.embeddingModelVersion!==void 0&&!me.test(e.embeddingModelVersion))throw new x("BAD_REQUEST",'@lunora/ai/rag: `embeddingModelVersion` must match /^[A-Za-z0-9._-]{1,40}$/ (a short, stable tag like "bge-v1.5")');const y=e.chunk??(h=>te(h,s,p)),{textStore:f}=e,k=f?de:ce,v=e.embeddingModelVersion,E=h=>v===void 0?h:h===void 0?v:`${v}::${h}`;return h=>{let I;const N=typeof h.trace=="function"?h.trace:void 0,B=async t=>{I??=ge(e.embeddingModel,h.ai);const n=I,o=async d=>{const{embedding:r,providerMetadata:u,usage:m}=await ee({model:n,value:t});if(d!==void 0){const c=m.tokens;typeof c=="number"&&Number.isFinite(c)&&d.setAttribute("gen_ai.usage.input_tokens",c);const l=we(u);l!==void 0&&d.setAttribute("gen_ai.usage.cost",l)}return r};if(N===void 0)return o();const i=fe(n),a=typeof h.conversationId=="string"&&h.conversationId.length>0?h.conversationId:void 0;return N("ai.embed",(d,r)=>o(r),{"gen_ai.operation.name":"embeddings",...i===void 0?{}:{"gen_ai.request.model":i},...a===void 0?{}:{"gen_ai.conversation.id":a}})},$=t=>{if(t===void 0){if(e.requireNamespace)throw new x("BAD_REQUEST",`@lunora/ai/rag: index "${e.index}" requires a namespace (requireNamespace is set) — pass the tenant/shard key on index()/retrieve()/remove()`);e.allowSharedNamespace||pe(e.index)}},U=async(t,n)=>{const[o]=await h.vectors.getByIds(e.index,[_(n,t,0)],n),i=o?.metadata?.[R],a=o?.metadata?.[T];return{chunks:typeof a=="number"&&Number.isInteger(a)&&a>0?a:void 0,hash:typeof i=="string"?i:void 0}},j=async(t,n,o,i)=>{const a=Array.from({length:o-n},(d,r)=>_(i,t,n+r));a.length!==0&&(await h.vectors.deleteByIds(e.index,a,i),await f?.remove?.(a,{namespace:i}),await e.lexicalStore?.remove?.(a,{namespace:i}))},W=async t=>{if($(t.namespace),t.importance!==void 0&&(typeof t.importance!="number"||t.importance<0||t.importance>1))throw new x("BAD_REQUEST","@lunora/ai/rag: `importance` must be a number in [0, 1]");const n=E(t.namespace),o=await he(t.text),i=await U(t.id,n);if(i.hash===o&&i.chunks!==void 0)return{chunks:i.chunks,ids:Array.from({length:i.chunks},(r,u)=>_(n,t.id,u)),unchanged:!0};const a=y(t.text),d=a.map((r,u)=>_(n,t.id,u));if(a.length===0&&t.allowEmptySources===!1)throw new x("BAD_REQUEST",`@lunora/ai/rag: source "${t.id}" produced zero chunks — set allowEmptySources: true to allow this`);if(a.length>0){const r=a.map((u,m)=>({chunkIndex:m,id:d[m],sourceId:t.id,text:u}));f&&await f.put(r,{namespace:n}),e.lexicalStore&&await e.lexicalStore.index(r,{namespace:n})}return await re(a,ae,async(r,u)=>{const m=d[u],c={...t.metadata,[z]:u,[V]:t.id};f||(c[A]=r),t.importance!==void 0&&(c[M]=t.importance),u===0&&(c[R]=o,c[T]=a.length,v!==void 0&&(c[F]=v)),await h.vectors.upsert(e.index,{embed:B,id:m,input:r,metadata:c,namespace:n}),t.onChunk?.({chunkIndex:u,id:m,text:r,total:a.length})}),i.chunks!==void 0&&i.chunks>a.length&&await j(t.id,a.length,i.chunks,n),{chunks:a.length,ids:d,unchanged:!1}},H=async t=>{$(t.namespace);const n=E(t.namespace),o=(await U(t.id,n)).chunks??1;await j(t.id,0,o,n)},q=async(t,n)=>{const o=new Map;if(t.length===0)return o;if(f){const a=await f.getMany(t,{namespace:n});for(const[d,r]of t.entries()){const u=a[d];typeof u=="string"&&o.set(r,u)}return o}const i=await h.vectors.getByIds(e.index,t,n);for(const a of i){const d=a.metadata?.[A];typeof d=="string"&&o.set(a.id,d)}return o},L=async(t,n,o)=>{const i=n?.chunkContext?.before??0,a=n?.chunkContext?.after??0;if(i===0&&a===0)return t;if(!Number.isInteger(i)||i<0||!Number.isInteger(a)||a<0)throw new x("BAD_REQUEST","@lunora/ai/rag: `chunkContext.before`/`chunkContext.after` must be non-negative integers");const d=new Map(t.map(c=>[c.id,c.text])),r=new Set;for(const c of t)for(let l=-i;l<=a;l+=1){const w=c.chunkIndex+l,g=_(o,c.sourceId,w);l!==0&&w>=0&&!d.has(g)&&r.add(g)}const u=await q([...r],o),m=(c,l)=>{const w=_(o,c,l);return d.get(w)??u.get(w)};return t.map(c=>{const l=[];for(let w=-i;w<=a;w+=1){const g=w===0?c.text:m(c.sourceId,c.chunkIndex+w);g!==void 0&&l.push(g)}return{...c,text:l.join(`
+`)}})},X=t=>{if(typeof t=="string"){const n=e.filters?.[t];if(!n)throw new x("NOT_FOUND",`@lunora/ai/rag: unknown named filter "${t}" — must be one of the keys declared in RagConfig.filters`);return n.filter}return t},Y=(t,n)=>t.matches.map(o=>{const i=o.metadata??{},a=K(o.id,n),d=i[A],r=i[M],u=typeof r=="number"&&r>=0&&r<=1?r:1;return{chunkIndex:a.chunkIndex,id:o.id,importance:u,metadata:O(i),score:o.score*u,sourceId:a.sourceId,text:typeof d=="string"?d:""}}),Z=async(t,n)=>{if(!f)return t;const o=t.map(r=>r.id),[i,a]=await Promise.all([q(o,n),h.vectors.getByIds(e.index,o,n)]),d=new Map(a.map(r=>[r.id,r.metadata]));return t.flatMap(r=>{const u=i.get(r.id);if(u===void 0)return[];const m=d.get(r.id),c=m?.[M],l=typeof c=="number"&&c>=0&&c<=1?c:r.importance,w=(r.importance===0?0:r.score/r.importance)*l;return[{...r,importance:l,metadata:O(m)??r.metadata,score:w,text:u}]})},C=async(t,n)=>{$(n?.namespace);const o=E(n?.namespace),i=X(n?.filter),a=e.rlsFilter?await e.rlsFilter(h.auth):void 0,d=a?{...i,...a}:i,r=Math.min(n?.topK??b,k),u=await h.vectors.query(e.index,{embed:B,filter:d,input:t,namespace:o,returnMetadata:f?"indexed":"all",topK:r});let m=await Z(Y(u,o),o);const c=n?.minScore;if(c!==void 0&&(m=m.filter(g=>g.score>=c)),e.lexicalStore){const g=(await e.lexicalStore.search(t,{filter:d,namespace:o,topK:e.lexicalTopK??r})).map(S=>{const D=K(S.id,o);return{chunkIndex:D.chunkIndex,id:S.id,importance:1,metadata:void 0,score:S.score,sourceId:D.sourceId,text:S.text}});m=[...ne(m,g)]}m.sort((g,S)=>S.score-g.score),m=[...await L(m,n,o)];const l=[],w=new Set;for(const g of m)w.has(g.sourceId)||(w.add(g.sourceId),l.push({id:g.sourceId,metadata:g.metadata,weight:g.importance}));return n?.onRetrieve?.({matches:m.length,query:t}),{chunks:m,context:le(m),sources:l}};return{asTool:t=>G({description:t?.description??`Search the "${e.index}" knowledge base for passages relevant to a natural-language query.`,execute:async({query:n})=>C(n,{namespace:t?.namespace,topK:t?.topK}),inputSchema:J({properties:{query:{description:"The natural-language search query.",type:"string"}},required:["query"],type:"object"})}),index:W,remove:H,retrieve:C}}};export{Ie as default};

package/dist/packem_shared/{types.d-F_Q8R8L3.d.ts → types.d-BcLGTChd.d.mts}

@@ -1,4 +1,15 @@
 import { EmbeddingModel, LanguageModel } from 'ai';
+/**
+ * Which surface resolved the gateway. The two paths handle the auth token
+ * differently: a bring-your-own AI SDK provider sends it as the
+ * `cf-aig-authorization` header (`ResolvedAiGateway.headers`), but the Workers AI
+ * **binding** routes through the gateway using the account's own credentials and
+ * its native `gateway` option (Cloudflare `GatewayOptions`) has **no** field to
+ * carry an authorization token — so a token set for the binding path cannot be
+ * delivered and is warned about instead of silently dropped.
+ * @experimental
+ */
+type AiGatewayConsumer = "byo-provider" | "workers-ai-binding";
 /**
  * Project {@link AiGatewayMetadata} to a plain string-map of only its defined,
  * non-empty fields, or `undefined` when nothing is set. This is the shared source
@@ -56,7 +67,21 @@ interface ResolvedAiGateway {
 declare const AI_GATEWAY_ACCOUNT_ID_ENV = "LUNORA_AI_GATEWAY_ACCOUNT_ID";
 /** Env var naming the AI Gateway id (the gateway's slug). */
 declare const AI_GATEWAY_ID_ENV = "LUNORA_AI_GATEWAY_ID";
-/** Env var carrying the gateway's authentication token (only for authenticated gateways). */
+/**
+ * Env var carrying the gateway's authentication token (only for authenticated
+ * gateways).
+ *
+ * **Workers AI binding limitation.** This token is delivered only on the
+ * bring-your-own AI SDK provider path, as the `cf-aig-authorization` header (see
+ * {@link ResolvedAiGateway.headers}). The Workers AI **binding** (`ctx.ai` over
+ * `env.AI`) routes through the gateway with the account's own credentials and its
+ * native `gateway` option (Cloudflare's `GatewayOptions`: `id` / `cacheKey` /
+ * `metadata` / …) has no authorization field — so an *authenticated* gateway that
+ * requires a token cannot be reached on the binding path. Set this only for a BYO
+ * provider; for Workers AI, leave the gateway unauthenticated (or front it with a
+ * BYO provider). {@link resolveAiGateway} warns once per isolate if the token is
+ * set on the binding path.
+ */
 declare const AI_GATEWAY_TOKEN_ENV = "LUNORA_AI_GATEWAY_TOKEN";
 /**
  * Resolve the Cloudflare AI Gateway coordinates from the Worker `env`, or
@@ -67,9 +92,16 @@ declare const AI_GATEWAY_TOKEN_ENV = "LUNORA_AI_GATEWAY_TOKEN";
  *
  * Pass optional {@link AiGatewayMetadata} to fold a `cf-aig-metadata` correlation
  * header into `headers` — only its defined fields are sent.
+ *
+ * `consumer` names the surface resolving the gateway (default `"byo-provider"`).
+ * When it is `"workers-ai-binding"` and an {@link AI_GATEWAY_TOKEN_ENV} token is
+ * configured, this warns once per isolate: the binding path cannot carry the
+ * token (see {@link AI_GATEWAY_TOKEN_ENV}), so it would otherwise be dropped with
+ * no diagnostic and every `ctx.ai.model(...)` call would fail against an
+ * authenticated gateway while the token var reads as "configured".
  * @experimental
  */
-declare const resolveAiGateway: (env: Record<string, unknown>, metadata?: AiGatewayMetadata) => ResolvedAiGateway | undefined;
+declare const resolveAiGateway: (env: Record<string, unknown>, metadata?: AiGatewayMetadata, consumer?: AiGatewayConsumer) => ResolvedAiGateway | undefined;
 /**
  * Structural projection of the Cloudflare Workers `AI` binding (`env.AI`).
  * Declared locally so unit tests can pass a plain-object double and the real

package/dist/packem_shared/{types.d-F_Q8R8L3.d.mts → types.d-BcLGTChd.d.ts}

@@ -1,4 +1,15 @@
 import { EmbeddingModel, LanguageModel } from 'ai';
+/**
+ * Which surface resolved the gateway. The two paths handle the auth token
+ * differently: a bring-your-own AI SDK provider sends it as the
+ * `cf-aig-authorization` header (`ResolvedAiGateway.headers`), but the Workers AI
+ * **binding** routes through the gateway using the account's own credentials and
+ * its native `gateway` option (Cloudflare `GatewayOptions`) has **no** field to
+ * carry an authorization token — so a token set for the binding path cannot be
+ * delivered and is warned about instead of silently dropped.
+ * @experimental
+ */
+type AiGatewayConsumer = "byo-provider" | "workers-ai-binding";
 /**
  * Project {@link AiGatewayMetadata} to a plain string-map of only its defined,
  * non-empty fields, or `undefined` when nothing is set. This is the shared source
@@ -56,7 +67,21 @@ interface ResolvedAiGateway {
 declare const AI_GATEWAY_ACCOUNT_ID_ENV = "LUNORA_AI_GATEWAY_ACCOUNT_ID";
 /** Env var naming the AI Gateway id (the gateway's slug). */
 declare const AI_GATEWAY_ID_ENV = "LUNORA_AI_GATEWAY_ID";
-/** Env var carrying the gateway's authentication token (only for authenticated gateways). */
+/**
+ * Env var carrying the gateway's authentication token (only for authenticated
+ * gateways).
+ *
+ * **Workers AI binding limitation.** This token is delivered only on the
+ * bring-your-own AI SDK provider path, as the `cf-aig-authorization` header (see
+ * {@link ResolvedAiGateway.headers}). The Workers AI **binding** (`ctx.ai` over
+ * `env.AI`) routes through the gateway with the account's own credentials and its
+ * native `gateway` option (Cloudflare's `GatewayOptions`: `id` / `cacheKey` /
+ * `metadata` / …) has no authorization field — so an *authenticated* gateway that
+ * requires a token cannot be reached on the binding path. Set this only for a BYO
+ * provider; for Workers AI, leave the gateway unauthenticated (or front it with a
+ * BYO provider). {@link resolveAiGateway} warns once per isolate if the token is
+ * set on the binding path.
+ */
 declare const AI_GATEWAY_TOKEN_ENV = "LUNORA_AI_GATEWAY_TOKEN";
 /**
  * Resolve the Cloudflare AI Gateway coordinates from the Worker `env`, or
@@ -67,9 +92,16 @@ declare const AI_GATEWAY_TOKEN_ENV = "LUNORA_AI_GATEWAY_TOKEN";
  *
  * Pass optional {@link AiGatewayMetadata} to fold a `cf-aig-metadata` correlation
  * header into `headers` — only its defined fields are sent.
+ *
+ * `consumer` names the surface resolving the gateway (default `"byo-provider"`).
+ * When it is `"workers-ai-binding"` and an {@link AI_GATEWAY_TOKEN_ENV} token is
+ * configured, this warns once per isolate: the binding path cannot carry the
+ * token (see {@link AI_GATEWAY_TOKEN_ENV}), so it would otherwise be dropped with
+ * no diagnostic and every `ctx.ai.model(...)` call would fail against an
+ * authenticated gateway while the token var reads as "configured".
  * @experimental
  */
-declare const resolveAiGateway: (env: Record<string, unknown>, metadata?: AiGatewayMetadata) => ResolvedAiGateway | undefined;
+declare const resolveAiGateway: (env: Record<string, unknown>, metadata?: AiGatewayMetadata, consumer?: AiGatewayConsumer) => ResolvedAiGateway | undefined;
 /**
  * Structural projection of the Cloudflare Workers `AI` binding (`env.AI`).
  * Declared locally so unit tests can pass a plain-object double and the real

package/dist/rag/index.d.mts

@@ -1,5 +1,5 @@
 import { Tool } from 'ai';
-import { E as EmbeddingModelInput, a as LunoraAi } from "../packem_shared/types.d-F_Q8R8L3.mjs";
+import { E as EmbeddingModelInput, a as LunoraAi } from "../packem_shared/types.d-BcLGTChd.mjs";
 /**
  * Built-in fixed-window chunker: split into `size`-char windows overlapping by
  * `overlap` chars. Deliberately simple and deterministic — the zero-config

package/dist/rag/index.d.ts

@@ -1,5 +1,5 @@
 import { Tool } from 'ai';
-import { E as EmbeddingModelInput, a as LunoraAi } from "../packem_shared/types.d-F_Q8R8L3.js";
+import { E as EmbeddingModelInput, a as LunoraAi } from "../packem_shared/types.d-BcLGTChd.js";
 /**
  * Built-in fixed-window chunker: split into `size`-char windows overlapping by
  * `overlap` chars. Deliberately simple and deterministic — the zero-config

package/dist/rag/index.mjs

@@ -1 +1 @@
-import{default as r}from"../packem_shared/fixedWindowChunks-XJRXHEoz.mjs";import{default as a}from"../packem_shared/defineRag-BF3RwHZj.mjs";import{contentHash as s,guessMimeTypeFromExtension as d}from"../packem_shared/contentHash-BIn6ECP8.mjs";import{default as n}from"../packem_shared/hybridRank-U6PmGuz1.mjs";import{default as i}from"../packem_shared/bm25LexicalStore-RA9sesFC.mjs";export{i as bm25LexicalStore,s as contentHash,a as defineRag,r as fixedWindowChunks,d as guessMimeTypeFromExtension,n as hybridRank};
+import{default as r}from"../packem_shared/fixedWindowChunks-XJRXHEoz.mjs";import{default as a}from"../packem_shared/defineRag-DUOcnSgC.mjs";import{contentHash as s,guessMimeTypeFromExtension as d}from"../packem_shared/contentHash-BIn6ECP8.mjs";import{default as n}from"../packem_shared/hybridRank-U6PmGuz1.mjs";import{default as i}from"../packem_shared/bm25LexicalStore-RA9sesFC.mjs";export{i as bm25LexicalStore,s as contentHash,a as defineRag,r as fixedWindowChunks,d as guessMimeTypeFromExtension,n as hybridRank};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lunora/ai",
-  "version": "1.0.0-alpha.26",
+  "version": "1.0.0-alpha.28",
   "description": "Workers AI helper for Lunora: provider-agnostic AI SDK access from functions, Workers AI by default",
   "keywords": [
     "ai",

package/dist/packem_shared/AI_GATEWAY_ACCOUNT_ID_ENV-JzYYnjLP.mjs

@@ -1 +0,0 @@
-const r=(t,a)=>{const n=t[a];return typeof n=="string"&&n.length>0?n:void 0},c=t=>{if(t===void 0)return;const a={};return typeof t.functionPath=="string"&&t.functionPath.length>0&&(a.functionPath=t.functionPath),typeof t.traceId=="string"&&t.traceId.length>0&&(a.traceId=t.traceId),Object.keys(a).length>0?a:void 0},d=t=>{const a=c(t);return a===void 0?void 0:JSON.stringify(a)},s="LUNORA_AI_GATEWAY_ACCOUNT_ID",_="LUNORA_AI_GATEWAY_ID",I="LUNORA_AI_GATEWAY_TOKEN",u=(t,a)=>{const n=r(t,s),o=r(t,_);if(n===void 0||o===void 0)return;const e=r(t,I),i={};e!==void 0&&(i["cf-aig-authorization"]=`Bearer ${e}`);const A=d(a);return A!==void 0&&(i["cf-aig-metadata"]=A),{accountId:n,baseURL:`https://gateway.ai.cloudflare.com/v1/${n}/${o}`,gatewayId:o,headers:i}};export{s as AI_GATEWAY_ACCOUNT_ID_ENV,_ as AI_GATEWAY_ID_ENV,I as AI_GATEWAY_TOKEN_ENV,c as buildAiGatewayMetadataFields,u as resolveAiGateway};

package/dist/packem_shared/createAi-k_AcrXgH.mjs

@@ -1 +0,0 @@
-import{LunoraError as t}from"@lunora/errors";import{createWorkersAI as v}from"workers-ai-provider";import{buildAiGatewayMetadataFields as c,resolveAiGateway as A}from"./AI_GATEWAY_ACCOUNT_ID_ENV-JzYYnjLP.mjs";const I=(i,o,s)=>{const d=c(s);if(i!==void 0)return d!==void 0&&i.metadata===void 0?{...i,metadata:d}:i;if(o===void 0)return;const a=A(o,s);if(a!==void 0)return d===void 0?{id:a.gatewayId}:{id:a.gatewayId,metadata:d}},y=(i,o)=>v({binding:i,gateway:o}),h=i=>{const{binding:o,defaultEmbeddingModel:s,defaultModel:d,env:a,gateway:f,metadata:g,provider:m}=i;if(!m&&!o)throw new t("INTERNAL","@lunora/ai: createAi requires a `binding` (env.AI) or a pre-built `provider`");const u=I(f,a,g),n=m??y(o,u),p=e=>{if(e===void 0){if(!d)throw new t("INTERNAL","@lunora/ai: no model supplied and no `defaultModel` configured — pass a model id or an AI SDK model");return n(d)}return typeof e=="string"?n(e):e},w=e=>{const r=n.textEmbeddingModel;if(typeof r!="function")throw new t("INTERNAL","@lunora/ai: the Workers AI provider does not expose `textEmbeddingModel`; pass an AI SDK EmbeddingModel (e.g. from @ai-sdk/openai) to embed()");return r.call(n,e)};return{embeddingModel:e=>{if(typeof e=="object")return e;const r=e??s;if(!r)throw new t("INTERNAL","@lunora/ai: no embedding model supplied and no `defaultEmbeddingModel` configured — pass an embedding model id or an AI SDK EmbeddingModel");return w(r)},model:p,run:async(e,r,l)=>{if(!o)throw new t("INTERNAL","@lunora/ai: ai.run requires the `binding` (env.AI) — it is unavailable when only a custom `provider` was supplied");const b=u!==void 0&&l?.gateway===void 0?{...l,gateway:u}:l;return o.run(e,r,b)},workersai:n}};export{h as default};