@fugood/buttress-server 2.26.0-beta.16 → 2.26.0-beta.18

package/lib/services/ggml-decision.d.ts

@@ -0,0 +1,80 @@
+import { z } from 'zod';
+import type { ReadableStream } from 'node:stream/web';
+import type { DecisionResult, ServiceContext, EventStream, Expand } from '../types';
+export declare const decideProperty: z.ZodObject<{
+    request: z.ZodObject<{
+        state: z.ZodUnion<[z.ZodString, z.ZodArray<z.ZodUnknown, "many">, z.ZodRecord<z.ZodString, z.ZodUnknown>]>;
+        questions: z.ZodRecord<z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+        images: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+    }, "strip", z.ZodTypeAny, {
+        state: string | unknown[] | Record<string, unknown>;
+        questions: Record<string, Record<string, unknown>>;
+        images?: string[] | undefined;
+    }, {
+        state: string | unknown[] | Record<string, unknown>;
+        questions: Record<string, Record<string, unknown>>;
+        images?: string[] | undefined;
+    }>;
+}, "strip", z.ZodTypeAny, {
+    request: {
+        state: string | unknown[] | Record<string, unknown>;
+        questions: Record<string, Record<string, unknown>>;
+        images?: string[] | undefined;
+    };
+}, {
+    request: {
+        state: string | unknown[] | Record<string, unknown>;
+        questions: Record<string, Record<string, unknown>>;
+        images?: string[] | undefined;
+    };
+}>;
+export declare const schemas: {
+    initContext: z.ZodTuple<[z.ZodString, z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>], null>;
+    decide: z.ZodTuple<[z.ZodString, z.ZodObject<{
+        request: z.ZodObject<{
+            state: z.ZodUnion<[z.ZodString, z.ZodArray<z.ZodUnknown, "many">, z.ZodRecord<z.ZodString, z.ZodUnknown>]>;
+            questions: z.ZodRecord<z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+            images: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        }, "strip", z.ZodTypeAny, {
+            state: string | unknown[] | Record<string, unknown>;
+            questions: Record<string, Record<string, unknown>>;
+            images?: string[] | undefined;
+        }, {
+            state: string | unknown[] | Record<string, unknown>;
+            questions: Record<string, Record<string, unknown>>;
+            images?: string[] | undefined;
+        }>;
+    }, "strip", z.ZodTypeAny, {
+        request: {
+            state: string | unknown[] | Record<string, unknown>;
+            questions: Record<string, Record<string, unknown>>;
+            images?: string[] | undefined;
+        };
+    }, {
+        request: {
+            state: string | unknown[] | Record<string, unknown>;
+            questions: Record<string, Record<string, unknown>>;
+            images?: string[] | undefined;
+        };
+    }>], null>;
+    releaseContext: z.ZodTuple<[z.ZodString], null>;
+};
+export interface Service {
+    initContext: (ctx: ServiceContext, ...args: z.infer<typeof schemas.initContext>) => ReadableStream<Expand<EventStream>>;
+    decide: (ctx: ServiceContext, ...args: z.infer<typeof schemas.decide>) => Promise<DecisionResult>;
+    releaseContext: (ctx: ServiceContext, ...args: z.infer<typeof schemas.releaseContext>) => Promise<{
+        released: boolean;
+        alreadyReleased?: boolean;
+    }>;
+}
+/**
+ * Clients upload local images first (like STT audio) and send their names;
+ * data URLs pass through.
+ */
+export declare const resolveDecisionImages: (session: ServiceContext['session'], request: z.infer<typeof decideProperty>['request']) => {
+    state: string | unknown[] | Record<string, unknown>;
+    questions: Record<string, Record<string, unknown>>;
+    images: string[] | undefined;
+};
+declare const _default: Service;
+export default _default;

package/lib/services/ggml-llm.d.ts

@@ -1,6 +1,6 @@
 import { z } from 'zod';
 import { type Service as LlmService } from './create-llm-service';
-import type { ServiceContext } from '../types';
+import type { DecisionResult, ServiceContext } from '../types';
 export declare const schemas: {
     initContext: z.ZodTuple<[z.ZodString, z.ZodOptional<z.ZodAny>], null>;
     completion: z.ZodTuple<[z.ZodString, z.ZodOptional<z.ZodAny>], null>;
@@ -9,9 +9,37 @@ export declare const schemas: {
     applyChatTemplate: z.ZodTuple<[z.ZodString, z.ZodAny], null>;
     releaseContext: z.ZodTuple<[z.ZodString], null>;
     embedding: z.ZodTuple<[z.ZodString, z.ZodAny], null>;
+    decide: z.ZodTuple<[z.ZodString, z.ZodObject<{
+        request: z.ZodObject<{
+            state: z.ZodUnion<[z.ZodString, z.ZodArray<z.ZodUnknown, "many">, z.ZodRecord<z.ZodString, z.ZodUnknown>]>;
+            questions: z.ZodRecord<z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+            images: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
+        }, "strip", z.ZodTypeAny, {
+            state: string | unknown[] | Record<string, unknown>;
+            questions: Record<string, Record<string, unknown>>;
+            images?: string[] | undefined;
+        }, {
+            state: string | unknown[] | Record<string, unknown>;
+            questions: Record<string, Record<string, unknown>>;
+            images?: string[] | undefined;
+        }>;
+    }, "strip", z.ZodTypeAny, {
+        request: {
+            state: string | unknown[] | Record<string, unknown>;
+            questions: Record<string, Record<string, unknown>>;
+            images?: string[] | undefined;
+        };
+    }, {
+        request: {
+            state: string | unknown[] | Record<string, unknown>;
+            questions: Record<string, Record<string, unknown>>;
+            images?: string[] | undefined;
+        };
+    }>], null>;
 };
 export interface Service extends LlmService {
     embedding: (ctx: ServiceContext, ...args: z.infer<typeof schemas.embedding>) => Promise<any>;
+    decide: (ctx: ServiceContext, ...args: z.infer<typeof schemas.decide>) => Promise<DecisionResult>;
 }
 declare const _default: Service;
 export default _default;

package/lib/services/index.d.ts

@@ -3,6 +3,7 @@ import type { Service as CommonService } from './common';
 import type { Service as GgmlLlmService } from './ggml-llm';
 import type { Service as GgmlSttService } from './ggml-stt';
 import type { Service as GgmlTtsService } from './ggml-tts';
+import type { Service as GgmlDecisionService } from './ggml-decision';
 import type { Service as MlxLlmService } from './mlx-llm';
 import type { Service as OnnxSttService } from './onnx-stt';
 import type { Service as OnnxTtsService } from './onnx-tts';
@@ -21,6 +22,7 @@ declare const services: {
         releaseContext: ({ backend, session }: ServiceContext, id: any, force: any) => Promise<any>;
     };
     ggmlTts: GgmlTtsService;
+    ggmlDecision: GgmlDecisionService;
     mlxLlm: MlxLlmService;
     onnxStt: OnnxSttService;
     onnxTts: OnnxTtsService;
@@ -35,6 +37,7 @@ export type Services = {
     ggmlLlm: ClientService<GgmlLlmService>;
     ggmlStt: ClientService<GgmlSttService>;
     ggmlTts: ClientService<GgmlTtsService>;
+    ggmlDecision: ClientService<GgmlDecisionService>;
     mlxLlm: ClientService<MlxLlmService>;
     onnxStt: ClientService<OnnxSttService>;
     onnxTts: ClientService<OnnxTtsService>;

package/lib/{tui-DbQ0zW-C.mjs → tui-Bxeb1d7E.mjs}

@@ -1,2 +1,2 @@
 #!/usr/bin/env node
-import{a as e,t}from"./client-DzfRSFcJ.mjs";import{Container as n,Editor as r,Loader as i,Markdown as a,ProcessTerminal as o,Text as s,TuiMainScreen as c}from"@earendil-works/pi-tui";const l=e=>t=>`\x1b[${e}m${t}\x1b[0m`,u=l(`2`),d=l(`1`),f=l(`36`),p=l(`33`),m=l(`31`),h=l(`32`),g=l(`35`),_={heading:e=>d(f(e)),link:f,linkUrl:e=>u(f(e)),code:p,codeBlock:p,codeBlockBorder:u,quote:u,quoteBorder:u,hr:u,listBullet:f,bold:d,italic:l(`3`),strikethrough:l(`9`),underline:l(`4`)},v={borderColor:u,selectList:{selectedPrefix:f,selectedText:d,description:u,scrollInfo:u,noMatch:u}},y=(e,t)=>e.length>t?`${e.slice(0,t)}…`:e;var b=class{tui;chat;thinking=null;thinkingBuffer=``;answer=null;answerBuffer=``;constructor(e,t){this.tui=e,this.chat=t}touch(){this.tui.requestRender()}appendThinking(e){this.thinking||(this.thinking=new s(``,0,0),this.chat.addChild(this.thinking)),this.thinkingBuffer+=e,this.thinking.setText(u(this.thinkingBuffer.trimEnd())),this.touch()}appendAnswer(e){this.answer||(this.answer=new a(``,0,0,_),this.chat.addChild(this.answer)),this.answerBuffer+=e,this.answer.setText(this.answerBuffer),this.touch()}addLine(e){this.answer=null,this.answerBuffer=``,this.chat.addChild(new s(e,0,0)),this.touch()}};const x=async({connection:a,agentName:l,sessionId:_,fork:x})=>{let S=new o,C=new c(S);S.setTitle(`buttress agent · ${l}`);let w=new n,T=new s(``,0,0),E=new r(C,v);C.addChild(w),C.addChild(T),C.addChild(E),C.setFocus(E);let D=_,O=x,k=null,A=null,j=e=>{T.setText(e),C.requestRender()},M=()=>{A||(A=new i(C,f,u,`thinking…`),w.addChild(A),C.requestRender())},N=()=>{A&&(w.removeChild(A),A=null,C.requestRender())},P=`${d(l)} ${u(`@ ${a.baseUrl}`)}`+(D?` ${u(`(${O?`forking`:`continuing`} ${D})`)}`:``);w.addChild(new s(P,0,0)),j(u(`Enter to send · /new fresh session · /exit or Ctrl+C to quit`));let F=async()=>{C.stop(),await S.drainInput().catch(()=>{}),process.exit(0)};C.addInputListener(e=>{if(e===``)return k?(k.abort(),k=null,{consume:!0}):(F(),{consume:!0})});let I=async n=>{w.addChild(new s(`${f(`>`)} ${n}`,0,0));let r=new b(C,w),i=D;k=new AbortController,M(),j(u(`running · Ctrl+C aborts this turn`));let o;try{o=await e(a,l,{prompt:n,sessionId:D??void 0,fork:O},{signal:k.signal,onFrame:({event:e,payload:t})=>{if(e===`session`&&t?.sessionId){t.sessionId!==D&&(D=t.sessionId,w.addChild(new s(u(`session ${D}`),0,0)));return}if(e!==`agent`)return;N();let n=t?.event;if(t?.type===`message_update`&&n)n.type===`thinking_delta`?r.appendThinking(n.delta??``):n.type===`text_delta`&&r.appendAnswer(n.delta??``);else if(t?.type===`tool_execution_start`){let e=JSON.stringify(t.args??{});r.addLine(`${p(`⚙ ${t.toolName}`)}${u(`(${y(e,120)})`)}`)}else t?.type===`tool_execution_end`?r.addLine(t.isError?m(`✗ ${t.toolName} failed`):u(`✓ ${t.toolName}`)):t?.type===`tool_emit`&&r.addLine(u(`  ${t.toolName} → ${t.event}`))}})}catch(e){o={kind:`error`,message:e?.message||String(e),sessionId:D}}if(N(),k=null,o.kind===`result`){let{result:e}=o;D=e.sessionId??D;let t=e.usage||{};w.addChild(new s(u(`— ${e.stopReason} · ${t.totalTurns??`?`} turn(s) · ${t.input??0} in / ${t.output??0} out · session ${D}`),0,0))}else o.kind===`aborted`?(D=o.sessionId??D,w.addChild(new s(p(`(aborted — session is continuable)`),0,0))):o.kind===`error`?(D=o.sessionId??D,w.addChild(new s(m(`Run failed: ${o.message}`),0,0))):w.addChild(new s(m(`Connection lost mid-run.`),0,0));t(i,D)&&(O=!1),j(u(`Enter to send · /new fresh session · /exit or Ctrl+C to quit`)),C.requestRender()};E.onSubmit=e=>{let t=e.trim();if(E.setText(``),t){if(k){j(p(`A turn is still running — Ctrl+C aborts it.`));return}if(t===`/exit`||t===`/quit`){F();return}if(t===`/new`){D=null,O=!1,w.addChild(new s(`${h(`•`)} ${u(`started a fresh session`)}`,0,0)),C.requestRender();return}if(t===`/session`){w.addChild(new s(u(D?`session ${D}`:`no session yet`),0,0)),C.requestRender();return}I(t)}},w.addChild(new s(u(g(`ready`)),0,0)),C.start(),await new Promise(()=>{})};export{x as runAgentTui};
+import{a as e,t}from"./client-BVyIuSyr.mjs";import{Container as n,Editor as r,Loader as i,Markdown as a,ProcessTerminal as o,Text as s,TuiMainScreen as c}from"@earendil-works/pi-tui";const l=e=>t=>`\x1b[${e}m${t}\x1b[0m`,u=l(`2`),d=l(`1`),f=l(`36`),p=l(`33`),m=l(`31`),h=l(`32`),g=l(`35`),_={heading:e=>d(f(e)),link:f,linkUrl:e=>u(f(e)),code:p,codeBlock:p,codeBlockBorder:u,quote:u,quoteBorder:u,hr:u,listBullet:f,bold:d,italic:l(`3`),strikethrough:l(`9`),underline:l(`4`)},v={borderColor:u,selectList:{selectedPrefix:f,selectedText:d,description:u,scrollInfo:u,noMatch:u}},y=(e,t)=>e.length>t?`${e.slice(0,t)}…`:e;var b=class{tui;chat;thinking=null;thinkingBuffer=``;answer=null;answerBuffer=``;constructor(e,t){this.tui=e,this.chat=t}touch(){this.tui.requestRender()}appendThinking(e){this.thinking||(this.thinking=new s(``,0,0),this.chat.addChild(this.thinking)),this.thinkingBuffer+=e,this.thinking.setText(u(this.thinkingBuffer.trimEnd())),this.touch()}appendAnswer(e){this.answer||(this.answer=new a(``,0,0,_),this.chat.addChild(this.answer)),this.answerBuffer+=e,this.answer.setText(this.answerBuffer),this.touch()}addLine(e){this.answer=null,this.answerBuffer=``,this.chat.addChild(new s(e,0,0)),this.touch()}};const x=async({connection:a,agentName:l,sessionId:_,fork:x})=>{let S=new o,C=new c(S);S.setTitle(`buttress agent · ${l}`);let w=new n,T=new s(``,0,0),E=new r(C,v);C.addChild(w),C.addChild(T),C.addChild(E),C.setFocus(E);let D=_,O=x,k=null,A=null,j=e=>{T.setText(e),C.requestRender()},M=()=>{A||(A=new i(C,f,u,`thinking…`),w.addChild(A),C.requestRender())},N=()=>{A&&(w.removeChild(A),A=null,C.requestRender())},P=`${d(l)} ${u(`@ ${a.baseUrl}`)}`+(D?` ${u(`(${O?`forking`:`continuing`} ${D})`)}`:``);w.addChild(new s(P,0,0)),j(u(`Enter to send · /new fresh session · /exit or Ctrl+C to quit`));let F=async()=>{C.stop(),await S.drainInput().catch(()=>{}),process.exit(0)};C.addInputListener(e=>{if(e===``)return k?(k.abort(),k=null,{consume:!0}):(F(),{consume:!0})});let I=async n=>{w.addChild(new s(`${f(`>`)} ${n}`,0,0));let r=new b(C,w),i=D;k=new AbortController,M(),j(u(`running · Ctrl+C aborts this turn`));let o;try{o=await e(a,l,{prompt:n,sessionId:D??void 0,fork:O},{signal:k.signal,onFrame:({event:e,payload:t})=>{if(e===`session`&&t?.sessionId){t.sessionId!==D&&(D=t.sessionId,w.addChild(new s(u(`session ${D}`),0,0)));return}if(e!==`agent`)return;N();let n=t?.event;if(t?.type===`message_update`&&n)n.type===`thinking_delta`?r.appendThinking(n.delta??``):n.type===`text_delta`&&r.appendAnswer(n.delta??``);else if(t?.type===`tool_execution_start`){let e=JSON.stringify(t.args??{});r.addLine(`${p(`⚙ ${t.toolName}`)}${u(`(${y(e,120)})`)}`)}else t?.type===`tool_execution_end`?r.addLine(t.isError?m(`✗ ${t.toolName} failed`):u(`✓ ${t.toolName}`)):t?.type===`tool_emit`&&r.addLine(u(`  ${t.toolName} → ${t.event}`))}})}catch(e){o={kind:`error`,message:e?.message||String(e),sessionId:D}}if(N(),k=null,o.kind===`result`){let{result:e}=o;D=e.sessionId??D;let t=e.usage||{};w.addChild(new s(u(`— ${e.stopReason} · ${t.totalTurns??`?`} turn(s) · ${t.input??0} in / ${t.output??0} out · session ${D}`),0,0))}else o.kind===`aborted`?(D=o.sessionId??D,w.addChild(new s(p(`(aborted — session is continuable)`),0,0))):o.kind===`error`?(D=o.sessionId??D,w.addChild(new s(m(`Run failed: ${o.message}`),0,0))):w.addChild(new s(m(`Connection lost mid-run.`),0,0));t(i,D)&&(O=!1),j(u(`Enter to send · /new fresh session · /exit or Ctrl+C to quit`)),C.requestRender()};E.onSubmit=e=>{let t=e.trim();if(E.setText(``),t){if(k){j(p(`A turn is still running — Ctrl+C aborts it.`));return}if(t===`/exit`||t===`/quit`){F();return}if(t===`/new`){D=null,O=!1,w.addChild(new s(`${h(`•`)} ${u(`started a fresh session`)}`,0,0)),C.requestRender();return}if(t===`/session`){w.addChild(new s(u(D?`session ${D}`:`no session yet`),0,0)),C.requestRender();return}I(t)}},w.addChild(new s(u(g(`ready`)),0,0)),C.start(),await new Promise(()=>{})};export{x as runAgentTui};

package/lib/types.d.ts

@@ -26,7 +26,25 @@ export type RuntimeConfig = {
     huggingface_token?: string;
     session_cache?: SessionCacheConfig;
 } & Record<string, any>;
-export type GeneratorType = 'ggml-llm' | 'ggml-stt' | 'ggml-tts' | 'mlx-llm' | 'onnx-stt' | 'onnx-tts';
+export type GeneratorType = 'ggml-llm' | 'ggml-stt' | 'ggml-tts' | 'ggml-decision' | 'mlx-llm' | 'onnx-stt' | 'onnx-tts';
+/** A TypeSafe `/v1/systemone` request, answered by a `ggml-decision` generator. */
+export type DecisionRequest = {
+    /** A string, or any JSON object/array (given to the model as JSON text). */
+    state: unknown;
+    /** Question id → `{ type: 'choice' | 'score' | 'noul', instructions, criteria }`. */
+    questions: Record<string, Record<string, any>>;
+    /** Local file paths or data URLs; needs a model with an mmproj configured. */
+    images?: string[];
+};
+export type DecisionResult = {
+    model: string;
+    /** Question id → the answer for its type (`choice`, `score` or `noul`). */
+    answers: Record<string, any>;
+    usage: {
+        input_tokens: number;
+        output_tokens: number;
+    };
+};
 export type GeneratorConfig = {
     type: GeneratorType;
 } & Record<string, any>;
@@ -40,6 +58,12 @@ export type GlobalConfig = {
         enabled?: boolean;
         cors_allowed_origins?: string | string[];
     };
+    typesafe_compat?: {
+        enabled?: boolean;
+        cors_allowed_origins?: string | string[];
+        /** Request `model` names (e.g. `jev-latest`) mapped to a decision generator id. */
+        models?: Record<string, string>;
+    };
     functions?: {
         enabled?: boolean;
         /** Directory holding local function files, relative to the config file. */

package/lib/utils/cors.d.ts

@@ -56,6 +56,7 @@ export type CorsPolicy = {
 export declare const buildCorsPolicies: (config: Config, enabled?: {
     openaiCompat?: boolean;
     anthropicMessages?: boolean;
+    typesafeCompat?: boolean;
     functions?: boolean;
 }) => CorsPolicy[];
 /**

package/lib/utils/decision-sharing.d.ts

@@ -0,0 +1,37 @@
+/**
+ * Decision models that share an LLM generator's context.
+ *
+ * A model that both generates and answers typed decisions (Jevling, for one)
+ * is often configured twice: as a `ggml-llm` and as a `ggml-decision`. With
+ * `[generators.backend] share_llm_context = true` on the decision entry, and
+ * both entries naming the same weights, decisions run on the LLM's context
+ * (`parallel.decide()` on a slot beside its completions) instead of loading
+ * the weights a second time. The LLM config always loads the model, so the
+ * result does not depend on which generator starts first.
+ *
+ * It is opt-in because it trades throughput for memory: a decision takes one
+ * of the LLM's `n_parallel` slots, and its forward passes run in the slot
+ * loop between two shared batches, so concurrent completions stop generating
+ * while it runs (a 128-token completion took 2x as long beside six decisions,
+ * unchanged with a separate decision context). Results are not affected: the
+ * slot's sequence and prompt cache are cleared after each decision.
+ *
+ * Sharing is skipped when the decision needs a context the LLM cannot give:
+ * rank pooling (an embedding context), or a different projector.
+ */
+/**
+ * The configured `ggml-llm` generator whose context serves this configured
+ * decision generator, or null. For a client's request use
+ * `findSharedLlmForClient`: the opt-in must come from the server config.
+ */
+export declare const findSharedLlmGenerator: (config: any, decisionConfig: any) => any | null;
+/**
+ * The shared LLM for a decision generator a client starts over RPC, or null.
+ * Sharing needs a configured `ggml-decision` entry with the same weights that
+ * opted in; the client's own `share_llm_context` is ignored, since it would
+ * let a client put its decisions on the LLM's slots. The client's merged
+ * model (rank pooling, projector) still decides whether the LLM can serve it.
+ */
+export declare const findSharedLlmForClient: (config: any, clientModel: any) => any | null;
+/** Startup report: which decision generators share an LLM, and why others do not. */
+export declare const describeDecisionSharing: (config: any) => string[];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fugood/buttress-server",
-  "version": "2.26.0-beta.16",
+  "version": "2.26.0-beta.18",
   "main": "lib/index.mjs",
   "types": "lib/index.d.ts",
   "type": "module",
@@ -19,7 +19,7 @@
   ],
   "scripts": {
     "typecheck": "tsc --noEmit",
-    "build": "tsdown -c rolldown.config.js --config-loader native && tsc --noCheck --emitDeclarationOnly && cp ../buttress-backend-core/src/backends/mlx-bridge.py lib/mlx-bridge.py",
+    "build": "tsdown -c rolldown.config.js --config-loader native && tsc --noCheck --emitDeclarationOnly && cp ../buttress-backend-core/src/backends/mlx-bridge.py lib/mlx-bridge.py && cp ../../node_modules/quickjs-wasi/quickjs.wasm lib/quickjs.wasm",
     "build:dist": "$npm_execpath scripts/build-distribution.js",
     "release": "$npm_execpath scripts/build-distribution.js",
     "release-beta": "$npm_execpath scripts/build-distribution.js --beta",
@@ -37,12 +37,12 @@
   ],
   "license": "MIT",
   "dependencies": {
-    "@earendil-works/pi-agent-core": "^0.99.1",
-    "@earendil-works/pi-ai": "0.99.1",
-    "@earendil-works/pi-tui": "^0.99.1",
+    "@earendil-works/pi-agent-core": "^1.0.0",
+    "@earendil-works/pi-ai": "1.0.0",
+    "@earendil-works/pi-tui": "^1.0.0",
     "@elysiajs/cors": "^1.1.1",
     "@elysiajs/node": "^1.4.2",
-    "@fugood/llama.node": "^1.8.0-rc.8",
+    "@fugood/llama.node": "^1.8.0-rc.9",
     "@fugood/whisper.node": "^1.2.0-rc.0",
     "@huggingface/gguf": "^0.3.2",
     "@iarna/toml": "^3.0.0",
@@ -69,10 +69,11 @@
     "zod": "^3.25.76"
   },
   "devDependencies": {
-    "@fugood/bricks-cli": "^2.26.0-beta.16",
-    "@fugood/bricks-devtools-core": "^2.26.0-beta.11",
+    "@fugood/agent-tool-runtime": "^2.26.0-beta.18",
+    "@fugood/bricks-cli": "^2.26.0-beta.18",
+    "@fugood/bricks-devtools-core": "^2.26.0-beta.18",
     "tsdown": "^0.22.4",
     "typescript": "^7.0.2"
   },
-  "gitHead": "c9cc188d85392d229bb7b143f9bf56e0d6f1cb23"
+  "gitHead": "56a905825a9905fbcf48a3ee7dcc689759b7f1f4"
 }

package/public/status.html

@@ -603,6 +603,27 @@
         </div>
       </div>
 
+      <!-- GGML Typed Decision Status (hidden unless active) -->
+      <div class="card" id="ggmlDecisionCard" style="display:none">
+        <div class="card-header">
+          <span class="card-title">Typed Decision Status</span>
+          <span class="badge badge-info" id="ggmlDecisionGeneratorCount">0 generators</span>
+        </div>
+        <div id="ggmlDecisionGenerators"></div>
+        <div class="section">
+          <div class="section-title collapsible" onclick="toggleSection(this)">Model Load History</div>
+          <div class="collapsible-content" id="ggmlDecisionModelHistory">
+            <div class="empty-state">No model load history</div>
+          </div>
+        </div>
+        <div class="section">
+          <div class="section-title collapsible" onclick="toggleSection(this)">Decision History</div>
+          <div class="collapsible-content" id="ggmlDecisionHistory">
+            <div class="empty-state">No decision history</div>
+          </div>
+        </div>
+      </div>
+
       <!-- Local Functions (hidden unless enabled) -->
       <div class="card" id="functionsCard" style="display:none">
         <div class="card-header">
@@ -1049,6 +1070,54 @@
       })
     }
 
+    function renderGgmlDecisionGenerators(generators) {
+      const container = document.getElementById('ggmlDecisionGenerators')
+      document.getElementById('ggmlDecisionGeneratorCount').textContent =
+        `${generators.length} generator${generators.length !== 1 ? 's' : ''}`
+
+      withScrollPreserve(container, () => {
+        if (generators.length === 0) {
+          container.innerHTML = '<div class="empty-state">No typed decision generators loaded</div>'
+          return
+        }
+
+        container.innerHTML = `
+          <div class="table-wrapper">
+            <div class="table-inner">
+              <table>
+                <thead>
+                  <tr>
+                    <th>ID</th>
+                    <th>Model</th>
+                    <th>Type</th>
+                    <th>Quant</th>
+                    <th>Variant</th>
+                    <th>GPU</th>
+                    <th>Model Size</th>
+                    <th>Context</th>
+                  </tr>
+                </thead>
+                <tbody>
+                  ${generators.map(g => `
+                    <tr>
+                      <td class="mono">${escapeHtml(g.id?.slice(0, 8) || '-')}</td>
+                      <td>${formatModelName(g.repoId)}${g.mmprojFilename ? '<br><span style="font-size:11px;color:var(--text-muted)">+ ' + escapeHtml(g.mmprojFilename) + '</span>' : ''}</td>
+                      <td>${escapeHtml(g.decisionType || '-')}</td>
+                      <td>${escapeHtml(g.quantization || '-')}</td>
+                      <td>${escapeHtml(g.variant || '-')}</td>
+                      <td><span class="badge ${g.useGpu ? 'badge-success' : 'badge-info'}">${g.useGpu ? 'yes' : 'no'}</span></td>
+                      <td>${formatBytes(g.modelBytes)}</td>
+                      <td>${g.hasContext ? '<span class="badge badge-success">loaded</span>' : '-'}</td>
+                    </tr>
+                  `).join('')}
+                </tbody>
+              </table>
+            </div>
+          </div>
+        `
+      })
+    }
+
     // Escape HTML
     function escapeHtml(str) {
       if (str == null) return ''
@@ -1077,6 +1146,7 @@
       const onnxStt = status.onnxStt || {}
       const onnxTts = status.onnxTts || {}
       const ggmlTts = status.ggmlTts || {}
+      const ggmlDecision = status.ggmlDecision || {}
 
       // LLM
       renderLlmGenerators(llm.generators || [])
@@ -1288,6 +1358,39 @@
         },
       ])
 
+      // GGML typed decisions
+      const hasGgmlDecision = (ggmlDecision.generators?.length > 0) ||
+        (ggmlDecision.history?.modelLoads?.length > 0) ||
+        (ggmlDecision.history?.decisions?.length > 0)
+      document.getElementById('ggmlDecisionCard').style.display = hasGgmlDecision ? '' : 'none'
+      renderGgmlDecisionGenerators(ggmlDecision.generators || [])
+
+      renderHistory('ggmlDecisionModelHistory', ggmlDecision.history?.modelLoads || [], [
+        { label: 'Time', render: i => `<span class="timestamp">${formatRelativeTime(i.timestamp)}</span>` },
+        { label: 'Model', render: i => formatModelName(i.repoId) },
+        { label: 'Type', render: i => escapeHtml(i.decisionType || '-') },
+        { label: 'Quant', render: i => escapeHtml(i.quantization || '-') },
+        { label: 'Variant', render: i => escapeHtml(i.variant || '-') },
+        { label: 'Duration', render: i => `${(i.durationMs / 1000).toFixed(1)}s` },
+        { label: 'Status', render: i => i.success ?
+          '<span class="badge badge-success">Success</span>' :
+          `<span class="badge badge-error">Failed: ${escapeHtml(i.error || 'Unknown')}</span>`
+        },
+      ])
+
+      renderHistory('ggmlDecisionHistory', ggmlDecision.history?.decisions || [], [
+        { label: 'Time', render: i => `<span class="timestamp">${formatRelativeTime(i.timestamp)}</span>` },
+        { label: 'Model', render: i => formatModelName(i.repoId) },
+        { label: 'Questions', render: i => i.questionCount ?? '-' },
+        { label: 'Images', render: i => i.imageCount || '-' },
+        { label: 'Tokens', render: i => i.inputTokens ?? '-' },
+        { label: 'Duration', render: i => `${i.durationMs} ms` },
+        { label: 'Status', render: i => i.success ?
+          '<span class="badge badge-success">Success</span>' :
+          `<span class="badge badge-error">Failed: ${escapeHtml(i.error || 'Unknown')}</span>`
+        },
+      ])
+
       renderFunctionsStatus(status.functions || {})
     }
 
@@ -1447,7 +1550,7 @@
         const response = await fetch('/buttress/status')
         const data = await response.json()
         // Direct JSON response (not tRPC wrapped)
-        if (data.ggmlLlm || data.ggmlStt || data.mlxLlm || data.onnxStt || data.onnxTts || data.ggmlTts) {
+        if (data.ggmlLlm || data.ggmlStt || data.mlxLlm || data.onnxStt || data.onnxTts || data.ggmlTts || data.ggmlDecision) {
           updateUI(data)
           setConnectionStatus('connected')
         }

package/lib/agent/session-fs.d.ts

@@ -1,3 +0,0 @@
-import type { FileSystem } from '@earendil-works/pi-agent-core';
-/** File storage for agent sessions that classifies errors structurally across vm realms. */
-export declare const createSessionFs: (rootDir: string) => FileSystem;

package/lib/config-DbRjQnNp.mjs

@@ -1,2 +0,0 @@
-#!/usr/bin/env node
-import e from"node:path";import t from"node:os";import n from"bytes";import r from"ms";import i from"node-machine-id";const a=(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]=a(n[e]||{},t):n[e]=t}),n},o=e=>e&&typeof e==`object`?structuredClone(e):null,s=(e,t)=>{let n=o(e)||{},r=o(t)||{};return a(n,r)},c=(e,t)=>a(structuredClone(e.global),t||{}),l=(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 c(e,a)}}return Object.keys(e.global).length>0?c(e,{}):null},u={udp:{port:8089,announcements:{enabled:!0,interval:5e3},requests:{enabled:!0,responseDelay:100}},http:{enabled:!0,path:`/buttress/info`,cors:!0}},d=e=>e?e===!0?{...u}:a(u,e):null,f=(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}))},p=(e,t,n)=>e===void 0?n:typeof e==`number`?e:t(e)??n,m=6e4,h=1024*1024*50,g=s=>{let c=i.machineIdSync(),l={server:{id:`buttress-${c}`,name:`Buttress Server (${c.slice(-8)})`,port:2080,temp_file_dir:e.join(t.tmpdir(),`.buttress`),session_timeout:m,max_body_size:h},autodiscover:!1},u=a(l,o(s)||{}),f=Array.isArray(u.generators)?u.generators:[],{server:g,generators:_,autodiscover:v,...y}=u;return{autodiscover:d(v),server:{id:g.id,name:g.name,port:g.port,log_level:g.log_level,temp_file_dir:g.temp_file_dir,max_body_size:p(g.max_body_size,n.parse,h),session_timeout:p(g.session_timeout,e=>r(e),m)},global:y,generators:f}},_=/^[A-Za-z0-9_-]{1,64}$/,v=n=>n===`~`||n.startsWith(`~/`)?e.join(t.homedir(),n.slice(1)):n;var y=class extends Error{constructor(e){super(`[Agents] ${e}`),this.name=`AgentsConfigError`}};const b=(e,t,n)=>{if(t==null)return n;if(t===0||t===!1)return null;if(typeof t==`number`){if(Number.isFinite(t)&&t>0)return t}else if(typeof t==`string`){let e;try{e=r(t)}catch{e=void 0}if(typeof e==`number`&&e>0)return e}throw new y(`agents_options.${e} must be a duration string (e.g. "30d") or a positive number of milliseconds; use 0 to disable it`)},x=(e,t)=>{if(t==null)return{};if(typeof t!=`object`||Array.isArray(t))throw new y(`agent "${e}": mcp_servers must be a table`);let n={};for(let[r,i]of Object.entries(t)){if(!_.test(r))throw new y(`agent "${e}": MCP server name "${r}" must match ${_}`);let t=!!(typeof i?.url==`string`&&i.url),a=!!(typeof i?.command==`string`&&i.command);if(t===a)throw new y(`agent "${e}": MCP server "${r}" needs exactly one of url / command`);n[r]={url:t?i.url:void 0,headers:i?.headers&&typeof i.headers==`object`?i.headers:void 0,command:a?i.command:void 0,args:Array.isArray(i?.args)?i.args.map(String):void 0,env:i?.env&&typeof i.env==`object`?i.env:void 0,optional:i?.optional===!0}}return n},S=(e,t,n)=>{let r=e?.name;if(typeof r!=`string`||!_.test(r))throw new y(`[[agents]] #${t+1}: name is required and must match ${_}`);let i=e?.model;if(typeof i!=`string`||!i.includes(`/`))throw new y(`agent "${r}": model must be "provider/model-id" (e.g. "buttress/ggml-org/gpt-oss-20b-GGUF")`);let a=i.indexOf(`/`),o=i.slice(0,a),s=i.slice(a+1);if(!o||!s)throw new y(`agent "${r}": invalid model reference "${i}"`);if(o===`buttress`&&!n.generators.filter(e=>e.type===`ggml-llm`||e.type===`mlx-llm`).some(e=>(e.model?.repo_id||e.model?.repository)===s))throw new y(`agent "${r}": model "${s}" does not match any configured [[generators]] ggml-llm/mlx-llm repo_id`);if(e?.system_prompt!=null&&e?.system_prompt_file!=null)throw new y(`agent "${r}": set system_prompt or system_prompt_file, not both`);let c=[];if(e?.tools!=null){if(!Array.isArray(e.tools))throw new y(`agent "${r}": tools must be an array of function names`);for(let t of e.tools){if(typeof t!=`string`||!_.test(t))throw new y(`agent "${r}": tool name "${t}" must match ${_} (local function names only)`);if(c.includes(t))throw new y(`agent "${r}": duplicate tool "${t}"`);c.push(t)}}if(e?.local_devices!==void 0&&typeof e.local_devices!=`boolean`)throw new y(`agent "${r}": local_devices must be a boolean`);let l=Number(e?.max_turns??30);if(!Number.isInteger(l)||l<1)throw new y(`agent "${r}": max_turns must be a positive integer`);let u=null;if(e?.max_tokens_per_run!=null&&(u=Number(e.max_tokens_per_run),!Number.isInteger(u)||u<1))throw new y(`agent "${r}": max_tokens_per_run must be a positive integer`);let{name:d,model:f,system_prompt:p,system_prompt_file:m,tools:h,local_devices:g=!1,mcp_servers:v,max_turns:b,max_tokens_per_run:S,thinking:C,...w}=e;return{name:r,provider:o,modelId:s,modelRef:i,systemPrompt:typeof p==`string`?p:null,systemPromptFile:typeof m==`string`?m:null,tools:c,localDevices:g,mcpServers:x(r,v),maxTurns:l,maxTokensPerRun:u,thinking:typeof C==`string`?C:void 0,generation:w}},C=(t,{configDir:n=process.cwd()}={})=>{let r=t.global||{},i=r.agents;if(i==null)return null;if(!Array.isArray(i))throw new y(`agents must be an array of tables ([[agents]])`);if(i.length===0)return null;let a=i.map((e,n)=>S(e,n,t)),o=new Set;for(let e of a){if(o.has(e.name))throw new y(`duplicate agent name "${e.name}"`);o.add(e.name)}let s=r.agents_options||{},c=typeof s.sessions_dir==`string`&&s.sessions_dir?s.sessions_dir:`./.buttress-agent/sessions`,l=v(c),u=e.isAbsolute(l)?l:e.resolve(n,l);for(let t of a)if(t.systemPromptFile){let r=v(t.systemPromptFile);t.systemPromptFile=e.isAbsolute(r)?r:e.resolve(n,r)}let d=Number(s.max_depth??2);if(!Number.isInteger(d)||d<1)throw new y(`agents_options.max_depth must be a positive integer`);let f=s.session_max_count,p=500;if(f===0||f===!1)p=null;else if(f!=null&&(p=Number(f),!Number.isInteger(p)||p<1))throw new y(`agents_options.session_max_count must be a positive integer`);return{agents:a,configDir:n,sessionsDir:u,runtimeTokenFile:e.join(e.dirname(u),`runtime-token`),sessionMaxAgeMs:b(`session_max_age`,s.session_max_age,2592e6),sessionMaxCount:p,maxDepth:d,allowUnauthenticated:s.allow_unauthenticated===!0}};export{g as a,o as i,C as n,l as o,s as r,f as s,y as t};