@mlx-node/agent 0.0.9 → 0.0.10

package/dist/provider/chat-config.d.ts

@@ -30,5 +30,5 @@ export interface ResolvedReasoningMode {
  * option was present.
  */
 export declare function resolveReasoningMode(reasoning: ThinkingLevel | undefined): ResolvedReasoningMode;
-export declare function buildChatConfig(modelType: ModelType, options: SimpleStreamOptions | undefined, tools: ToolDefinition[] | undefined, rootCacheOwnerId?: string, resolvedReasoning?: ResolvedReasoningMode): ChatConfig;
+export declare function buildChatConfig(modelType: ModelType, options: SimpleStreamOptions | undefined, tools: ToolDefinition[] | undefined, rootCacheOwnerId?: string, resolvedReasoning?: ResolvedReasoningMode, modelMaxTokens?: unknown): ChatConfig;
 //# sourceMappingURL=chat-config.d.ts.map

package/dist/provider/chat-config.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"chat-config.d.ts","sourceRoot":"","sources":["../../src/provider/chat-config.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,mBAAmB,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAChF,OAAO,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC1E,OAAO,EAAkB,KAAK,YAAY,EAAE,MAAM,kBAAkB,CAAC;AA0BrE;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,SAAS,EAAE,SAAS,GAAG,YAAY,GAAG,SAAS,CAE9E;AAgBD,MAAM,WAAW,qBAAqB;IACpC,eAAe,EAAE,MAAM,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IACpD,8EAA8E;IAC9E,eAAe,EAAE,OAAO,CAAC;CAC1B;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,SAAS,EAAE,aAAa,GAAG,SAAS,GAAG,qBAAqB,CAMhG;AAED,wBAAgB,eAAe,CAC7B,SAAS,EAAE,SAAS,EACpB,OAAO,EAAE,mBAAmB,GAAG,SAAS,EACxC,KAAK,EAAE,cAAc,EAAE,GAAG,SAAS,EACnC,gBAAgB,CAAC,EAAE,MAAM,EACzB,iBAAiB,wBAA2C,GAC3D,UAAU,CA6BZ"}
+{"version":3,"file":"chat-config.d.ts","sourceRoot":"","sources":["../../src/provider/chat-config.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,mBAAmB,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAChF,OAAO,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC1E,OAAO,EAAkB,KAAK,YAAY,EAAE,MAAM,kBAAkB,CAAC;AA0BrE;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,SAAS,EAAE,SAAS,GAAG,YAAY,GAAG,SAAS,CAE9E;AAgBD,MAAM,WAAW,qBAAqB;IACpC,eAAe,EAAE,MAAM,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IACpD,8EAA8E;IAC9E,eAAe,EAAE,OAAO,CAAC;CAC1B;AAMD;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,SAAS,EAAE,aAAa,GAAG,SAAS,GAAG,qBAAqB,CAMhG;AAED,wBAAgB,eAAe,CAC7B,SAAS,EAAE,SAAS,EACpB,OAAO,EAAE,mBAAmB,GAAG,SAAS,EACxC,KAAK,EAAE,cAAc,EAAE,GAAG,SAAS,EACnC,gBAAgB,CAAC,EAAE,MAAM,EACzB,iBAAiB,wBAA2C,EAC5D,cAAc,CAAC,EAAE,OAAO,GACvB,UAAU,CAsCZ"}

package/dist/provider/chat-config.js

@@ -54,6 +54,9 @@ const THINKING_LEVEL_TO_EFFORT = {
     xhigh: 'high',
     max: 'high',
 };
+function isPositiveSafeInteger(value) {
+    return typeof value === 'number' && Number.isSafeInteger(value) && value > 0;
+}
 /**
  * Resolve Pi's thinking level once for both native config and persisted replay
  * provenance. Keeping these values together prevents a low/minimal turn from
@@ -67,7 +70,7 @@ export function resolveReasoningMode(reasoning) {
         thinkingEnabled: reasoningEffort === 'medium' || reasoningEffort === 'high',
     };
 }
-export function buildChatConfig(modelType, options, tools, rootCacheOwnerId, resolvedReasoning = resolveReasoningMode(options?.reasoning)) {
+export function buildChatConfig(modelType, options, tools, rootCacheOwnerId, resolvedReasoning = resolveReasoningMode(options?.reasoning), modelMaxTokens) {
     const preset = launchPresetFor(modelType);
     if (!preset) {
         const known = [...new Set([...Object.keys(LAUNCH_PRESETS), ...Object.keys(AGENT_LAUNCH_PRESETS)])].join(', ');
@@ -92,8 +95,17 @@ export function buildChatConfig(modelType, options, tools, rootCacheOwnerId, res
     // which branch the bounded GDN sidecar store protects from child eviction.
     if (rootCacheOwnerId !== undefined)
         config.cacheRootOwnerId = rootCacheOwnerId;
-    if (options?.maxTokens !== undefined)
-        config.maxNewTokens = options.maxTokens;
+    const explicitMaxTokens = options?.maxTokens;
+    if (isPositiveSafeInteger(explicitMaxTokens)) {
+        // A valid per-call provider option is the topmost layer.
+        config.maxNewTokens = explicitMaxTokens;
+    }
+    else if (isPositiveSafeInteger(modelMaxTokens)) {
+        // Normal Pi agent turns omit SimpleStreamOptions.maxTokens. Honor the
+        // composed Model metadata (including models.json modelOverrides) without
+        // allowing malformed/hostile metadata to replace the family preset.
+        config.maxNewTokens = modelMaxTokens;
+    }
     if (options?.temperature !== undefined)
         config.temperature = options.temperature;
     if (tools && tools.length > 0)

package/dist/provider/stream-adapter.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"stream-adapter.d.ts","sourceRoot":"","sources":["../../src/provider/stream-adapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoCG;AAEH,OAAO,KAAK,EACV,GAAG,EACH,gBAAgB,EAChB,2BAA2B,EAC3B,OAAO,EACP,KAAK,EACL,mBAAmB,EACpB,MAAM,uBAAuB,CAAC;AAE/B,OAAO,KAAK,EAAE,WAAW,EAAE,eAAe,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAErF,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAOvD;;;;GAIG;AACH,MAAM,WAAW,gBAAgB;IAC/B,kFAAkF;IAClF,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,mBAAmB,GAAG,SAAS,CAAC;IAC5D;;;;OAIG;IACH,eAAe,CAAC,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC,OAAO,EAAE,WAAW,EAAE,QAAQ,EAAE,OAAO,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IAC7G,+FAA+F;IAC/F,iBAAiB,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACzC,oFAAoF;IACpF,oBAAoB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IAC/C,gFAAgF;IAChF,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3C;AAED,MAAM,MAAM,mBAAmB,GAAG,CAAC,OAAO,EAAE,gBAAgB,EAAE,WAAW,EAAE,kBAAkB,KAAK,IAAI,CAAC;AACvG,MAAM,MAAM,sBAAsB,GAAG,MAAM,MAAM,GAAG,SAAS,CAAC;AAC9D,qFAAqF;AACrF,MAAM,MAAM,uBAAuB,GAAG,MAAM,MAAM,GAAG,SAAS,CAAC;AAE/D;;;;;;GAMG;AACH,MAAM,MAAM,YAAY,GAAG,CAAC,GAAG,EAAE;IAC/B,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iFAAiF;IACjF,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,eAAe,CAAC;IACvB,sFAAsF;IACtF,UAAU,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB,8FAA8F;IAC9F,QAAQ,EAAE,OAAO,CAAC;CACnB,KAAK,IAAI,CAAC;AAqFX,wBAAgB,mBAAmB,CACjC,IAAI,EAAE,gBAAgB,EACtB,aAAa,CAAC,EAAE,mBAAmB,EACnC,qBAAqB,CAAC,EAAE,sBAAsB,EAC9C,YAAY,CAAC,EAAE,YAAY,EAC3B,WAAW,CAAC,EAAE,MAAM,IAAI,EACxB,sBAAsB,CAAC,EAAE,uBAAuB,GAC/C,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,EAAE,mBAAmB,KAAK,2BAA2B,CAuRrG"}
+{"version":3,"file":"stream-adapter.d.ts","sourceRoot":"","sources":["../../src/provider/stream-adapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoCG;AAEH,OAAO,KAAK,EACV,GAAG,EACH,gBAAgB,EAChB,2BAA2B,EAC3B,OAAO,EACP,KAAK,EACL,mBAAmB,EACpB,MAAM,uBAAuB,CAAC;AAE/B,OAAO,KAAK,EAAE,WAAW,EAAE,eAAe,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAErF,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAOvD;;;;GAIG;AACH,MAAM,WAAW,gBAAgB;IAC/B,kFAAkF;IAClF,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,mBAAmB,GAAG,SAAS,CAAC;IAC5D;;;;OAIG;IACH,eAAe,CAAC,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC,OAAO,EAAE,WAAW,EAAE,QAAQ,EAAE,OAAO,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IAC7G,+FAA+F;IAC/F,iBAAiB,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACzC,oFAAoF;IACpF,oBAAoB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC;IAC/C,gFAAgF;IAChF,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3C;AAED,MAAM,MAAM,mBAAmB,GAAG,CAAC,OAAO,EAAE,gBAAgB,EAAE,WAAW,EAAE,kBAAkB,KAAK,IAAI,CAAC;AACvG,MAAM,MAAM,sBAAsB,GAAG,MAAM,MAAM,GAAG,SAAS,CAAC;AAC9D,qFAAqF;AACrF,MAAM,MAAM,uBAAuB,GAAG,MAAM,MAAM,GAAG,SAAS,CAAC;AAE/D;;;;;;GAMG;AACH,MAAM,MAAM,YAAY,GAAG,CAAC,GAAG,EAAE;IAC/B,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iFAAiF;IACjF,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,eAAe,CAAC;IACvB,sFAAsF;IACtF,UAAU,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB,8FAA8F;IAC9F,QAAQ,EAAE,OAAO,CAAC;CACnB,KAAK,IAAI,CAAC;AAwGX,wBAAgB,mBAAmB,CACjC,IAAI,EAAE,gBAAgB,EACtB,aAAa,CAAC,EAAE,mBAAmB,EACnC,qBAAqB,CAAC,EAAE,sBAAsB,EAC9C,YAAY,CAAC,EAAE,YAAY,EAC3B,WAAW,CAAC,EAAE,MAAM,IAAI,EACxB,sBAAsB,CAAC,EAAE,uBAAuB,GAC/C,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,EAAE,mBAAmB,KAAK,2BAA2B,CA6RrG"}

package/dist/provider/stream-adapter.js

@@ -51,6 +51,16 @@ function safeString(read, fallback) {
         return fallback;
     }
 }
+/** Read numeric model metadata without trusting a user-configurable getter. */
+function readPositiveSafeInteger(read) {
+    try {
+        const value = read();
+        return typeof value === 'number' && Number.isSafeInteger(value) && value > 0 ? value : undefined;
+    }
+    catch {
+        return undefined;
+    }
+}
 /**
  * Publish the native model's load-time physical context limit onto pi's shared
  * model object. The parent session and every in-process subagent resolve this
@@ -62,13 +72,18 @@ function safeString(read, fallback) {
  * discovery-time limit here, and never let metadata synchronization break an
  * otherwise valid inference turn.
  */
-function publishEffectiveContextWindow(model, session) {
+function publishEffectiveContextWindow(model, session, configuredModelMaxTokens) {
     try {
         const effective = Math.floor(session.contextLimits()?.effectiveWindowTokens ?? 0);
         if (!Number.isSafeInteger(effective) || effective <= 0)
             return;
         model.contextWindow = Math.min(model.contextWindow, effective);
-        model.maxTokens = Math.min(model.maxTokens, model.contextWindow);
+        // Use the already-validated snapshot. Reading the shared metadata here
+        // would let Math.min coerce invalid user configuration into a valid value
+        // that a later turn would then trust.
+        if (configuredModelMaxTokens !== undefined) {
+            model.maxTokens = Math.min(configuredModelMaxTokens, model.contextWindow);
+        }
     }
     catch {
         // Exact native preflight still protects capacity; keep serving the turn.
@@ -272,7 +287,12 @@ export function makeMlxStreamSimple(host, onPerformance, resolveRootCacheOwner,
                         // Telemetry snapshot is best-effort; never break inference.
                     }
                 }
-                publishEffectiveContextWindow(model, session);
+                // Snapshot the composed budget before publishing native context limits.
+                // The advisory publisher must not read this raw value itself:
+                // Math.min can coerce hostile metadata (for example "512" or Infinity)
+                // into a finite number that a later turn would then trust.
+                const configuredModelMaxTokens = readPositiveSafeInteger(() => model.maxTokens);
+                publishEffectiveContextWindow(model, session, configuredModelMaxTokens);
                 const supportsImages = publishImageCapability(model, session);
                 const discovered = host.modelInfo(model.id);
                 if (!discovered) {
@@ -299,7 +319,7 @@ export function makeMlxStreamSimple(host, onPerformance, resolveRootCacheOwner,
                     return;
                 try {
                     session.primeHistory(contextToChatMessages(context, supportsImages));
-                    const config = buildChatConfig(discovered.modelType, options, toolsToDefinitions(context.tools), rootCacheOwnerId, resolvedReasoning);
+                    const config = buildChatConfig(discovered.modelType, options, toolsToDefinitions(context.tools), rootCacheOwnerId, resolvedReasoning, configuredModelMaxTokens);
                     for await (const event of session.startFromHistoryStream(config, signal)) {
                         if (event.done) {
                             if (event.finishReason === 'error') {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mlx-node/agent",
-  "version": "0.0.9",
+  "version": "0.0.10",
   "homepage": "https://github.com/mlx-node/mlx-node",
   "bugs": {
     "url": "https://github.com/mlx-node/mlx-node/issues"
@@ -33,9 +33,9 @@
   "dependencies": {
     "@earendil-works/pi-ai": "0.81.1",
     "@earendil-works/pi-coding-agent": "0.81.1",
-    "@mlx-node/core": "0.0.9",
-    "@mlx-node/lm": "0.0.9",
-    "@mlx-node/server": "0.0.9",
+    "@mlx-node/core": "0.0.10",
+    "@mlx-node/lm": "0.0.10",
+    "@mlx-node/server": "0.0.10",
     "typebox": "1.3.6"
   },
   "devDependencies": {