@kuralle-agents/core 0.3.11 → 0.3.13

package/dist/runtime/Runtime.d.ts

@@ -48,6 +48,7 @@ export declare class Runtime {
     private readonly terminalHandoffTargets;
     private readonly hooks?;
     private readonly activeTurnAborts;
+    private readonly sessionMutex;
     constructor(config: HarnessConfig);
     run(opts: RunOptions): TurnHandle;
     stream(opts: RunOptions): TurnHandle;

package/dist/runtime/Runtime.js

@@ -15,6 +15,7 @@ import { loadRecordedSteps } from './durable/replay.js';
 import { markSessionOutcome } from './outcomeMarking.js';
 import { resolveAgentPolicies } from './policies/resolvePolicies.js';
 import { buildAutoRetrieveProvider, buildKnowledgeProvider, buildMemoryService, runMemoryIngest, } from './grounding/index.js';
+import { SessionMutex } from './SessionMutex.js';
 export class Runtime {
     config;
     agentsById;
@@ -24,6 +25,7 @@ export class Runtime {
     terminalHandoffTargets;
     hooks;
     activeTurnAborts = new Map();
+    sessionMutex = new SessionMutex();
     constructor(config) {
         this.config = config;
         this.agentsById = indexAgents(config.agents);
@@ -89,6 +91,7 @@ export class Runtime {
                 steps,
                 toolExecutor,
                 model,
+                controlModel: opened.agent.controlModel ?? model,
                 abortSignal: abortController.signal,
                 emit,
                 refinementPolicies: policies.refinementPolicies,
@@ -201,10 +204,19 @@ export class Runtime {
             }
             return { text: collectAssistantText(runCtx.runState.messages), toolResults: [] };
         };
+        const gated = async () => {
+            const release = await this.sessionMutex.acquire(sessionId);
+            try {
+                return await execute();
+            }
+            finally {
+                release();
+            }
+        };
         return createTurnHandle({
             bus,
             abortController,
-            run: execute,
+            run: gated,
         });
     }
     stream(opts) {

package/dist/runtime/SessionMutex.js

@@ -47,8 +47,7 @@ export class SessionMutex {
         // Set the new tail BEFORE awaiting — so the next concurrent caller
         // sees this promise in the chain
         this.locks.set(sessionId, newTail);
-        // Wait for the previous lock holder to release
-        await currentTail;
+        await currentTail.catch(() => { });
         return releaseFn;
     }
     /** Number of sessions currently locked. For testing/debugging. */

package/dist/runtime/channels/TextDriver.js

@@ -109,8 +109,7 @@ export class TextDriver {
     // voice are identical). The model's prose is discarded; the user-facing
     // question is emitted deterministically by the flow engine (CollectNode.ask).
     runExtraction(node, ctx) {
-        const model = node.node.model ?? ctx.model;
-        return runSilentExtraction(node, ctx, model, resolveMaxSteps(ctx.limits, this.maxSteps));
+        return runSilentExtraction(node, ctx, ctx.controlModel, resolveMaxSteps(ctx.limits, this.maxSteps));
     }
     async runStructured(node, ctx) {
         const base = composeSystem(ctx.baseInstructions, resolveInstructions(node.instructions, ctx.runState.state), ctx.runState.state);
@@ -125,10 +124,11 @@ export class TextDriver {
                 .join(', ')}. Respond with only the chosen id, nothing else.`
             : base;
         const { object } = await generateObject({
-            model: ctx.model,
+            model: ctx.controlModel,
             schema,
             system,
             messages: ctx.runState.messages,
+            temperature: 0,
             abortSignal: ctx.abortSignal,
         });
         return object;

package/dist/runtime/channels/VoiceDriver.js

@@ -75,10 +75,11 @@ export class VoiceDriver {
                 .join(', ')}. Respond with only the chosen id, nothing else.`
             : base;
         const { object } = await generateObject({
-            model: ctx.model,
+            model: ctx.controlModel,
             schema,
             system,
             messages: ctx.runState.messages,
+            temperature: 0,
             abortSignal: ctx.abortSignal,
         });
         return object;
@@ -89,7 +90,7 @@ export class VoiceDriver {
     // behavior to TextDriver — voice and text emit the same structural events; the
     // user-facing question is the deterministic CollectNode.ask, synthesized after.
     runExtraction(node, ctx) {
-        return runSilentExtraction(node, ctx, ctx.model, resolveMaxSteps(ctx.limits, this.maxSteps));
+        return runSilentExtraction(node, ctx, ctx.controlModel, resolveMaxSteps(ctx.limits, this.maxSteps));
     }
     async awaitUser(ctx) {
         if (this.pendingBargeInInput != null) {

package/dist/runtime/channels/extractionTurn.js

@@ -20,7 +20,14 @@ export async function runSilentExtraction(node, ctx, model, maxSteps) {
     const aiTools = resolveExtractionTools(node);
     const out = { text: '', toolResults: [] };
     for (let step = 0; step < maxSteps; step += 1) {
-        const result = streamText({ model, system, messages, tools: aiTools, abortSignal: ctx.abortSignal });
+        const result = streamText({
+            model,
+            system,
+            messages,
+            tools: aiTools,
+            temperature: 0,
+            abortSignal: ctx.abortSignal,
+        });
         for await (const part of result.fullStream) {
             // Intentionally NOT handling 'text-delta' — extraction never speaks.
             if (part.type === 'error') {

package/dist/runtime/channels/inputBuffer.js

@@ -1,19 +1,27 @@
 const PENDING_INPUT_KEY = '__v2_pendingUserInput';
+function queue(session) {
+    const v = session.workingMemory[PENDING_INPUT_KEY];
+    if (Array.isArray(v))
+        return v;
+    if (typeof v === 'string' && v.length > 0)
+        return [v];
+    return [];
+}
 export function setPendingUserInput(session, input) {
-    session.workingMemory[PENDING_INPUT_KEY] = input;
+    session.workingMemory[PENDING_INPUT_KEY] = [...queue(session), input];
 }
 export function consumePendingUserInput(session) {
-    const value = session.workingMemory[PENDING_INPUT_KEY];
-    delete session.workingMemory[PENDING_INPUT_KEY];
-    if (typeof value !== 'string' || value.length === 0) {
-        throw new Error('No buffered user input for awaitUser');
-    }
-    return value;
+    const q = queue(session);
+    const next = q.shift() ?? '';
+    if (q.length === 0)
+        delete session.workingMemory[PENDING_INPUT_KEY];
+    else
+        session.workingMemory[PENDING_INPUT_KEY] = q;
+    return next;
 }
 export function peekPendingUserInput(session) {
-    const value = session.workingMemory[PENDING_INPUT_KEY];
-    return typeof value === 'string' ? value : undefined;
+    return queue(session)[0];
 }
 export function hasPendingUserInput(session) {
-    return peekPendingUserInput(session) !== undefined;
+    return queue(session).length > 0;
 }

package/dist/runtime/ctx.d.ts

@@ -20,6 +20,7 @@ export interface CtxDeps {
     toolExecutor: EffectToolExecutor;
     hookRunner?: HookRunner;
     model: LanguageModel;
+    controlModel?: LanguageModel;
     refinementPolicies?: RefinementCapability[];
     validationPolicies?: ValidationCapability[];
     inputProcessors?: InputProcessor[];

package/dist/runtime/ctx.js

@@ -99,6 +99,7 @@ function makeCtx(deps) {
         toolExecutor: deps.toolExecutor,
         hookRunner: deps.hookRunner ?? {},
         model: deps.model,
+        controlModel: deps.controlModel ?? deps.model,
         refinementPolicies: deps.refinementPolicies ?? [],
         validationPolicies: deps.validationPolicies ?? [],
         inputProcessors: deps.inputProcessors ?? [],

package/dist/runtime/hostLoop.js

@@ -21,7 +21,7 @@ export async function hostLoop(options) {
             const selection = await select({
                 agent,
                 run,
-                model: agent.routing?.model ?? agent.model ?? ctx.model,
+                model: agent.routing?.model ?? ctx.controlModel,
                 alwaysRoute,
             });
             if (selection.kind === 'enterFlow') {

package/dist/runtime/select.js

@@ -33,6 +33,7 @@ export async function selectHostTarget(options) {
     const { object } = await generateObject({
         model,
         schema: selectionSchema,
+        temperature: 0,
         system: 'You are an internal routing classifier. Choose exactly one action. ' +
             'Output schema fields only — never user-facing prose. ' +
             'Prefer route when a route condition clearly matches; else enterFlow when a flow description matches an uncompleted flow; else keep. Never re-enter a completed flow.',

package/dist/types/agentConfig.d.ts

@@ -14,6 +14,10 @@ export interface AgentConfig {
     description?: string;
     instructions?: Instructions;
     model?: LanguageModel;
+    /** Optional model for the control path (routing, decide, extraction), run at
+     *  temperature 0 for determinism. Defaults to `model` (the speaker) when unset.
+     *  Set this to pin control to a reliable provider independent of the speaker. */
+    controlModel?: LanguageModel;
     tools?: ToolSet;
     effectTools?: Record<string, AnyTool>;
     /** Safe, always-available tools made model-visible in EVERY speaking node turn

package/dist/types/run-context.d.ts

@@ -51,6 +51,8 @@ export interface RunContext {
     toolExecutor: EffectToolExecutor;
     hookRunner: HookRunner;
     model: LanguageModel;
+    /** Control-path model (routing, decide, extraction) at temperature 0. */
+    controlModel: LanguageModel;
     refinementPolicies: RefinementCapability[];
     validationPolicies: ValidationCapability[];
     inputProcessors: InputProcessor[];

package/package.json

@@ -6,7 +6,7 @@
     "url": "git+https://github.com/kuralle/kuralle-agents.git",
     "directory": "packages/kuralle-core"
   },
-  "version": "0.3.11",
+  "version": "0.3.13",
   "description": "A framework for structured conversational AI agents",
   "publishConfig": {
     "access": "public"
@@ -97,7 +97,7 @@
     "dotenv": "^16.4.0",
     "typescript": "^5.3.0",
     "zod": "^3.23.0",
-    "@kuralle-agents/realtime-audio": "0.3.11"
+    "@kuralle-agents/realtime-audio": "0.3.13"
   },
   "dependencies": {
     "chrono-node": "^2.6.0",