@fonoster/autopilot 0.30.0 → 0.31.0

package/README.md

@@ -15,7 +15,7 @@ The autopilot module provides a way to create Voice AI applications in Fonoster.
 
 ## Installation
 
-Autopilot is an ordinary session app that the VoiceSessionGateway dials. Run the
+Autopilot is an ordinary session app that AppBridge dials. Run the
 `fonoster/autopilot` image (built from `mods/autopilot/Dockerfile`), or from the repository:
 
 ```sh-session
@@ -23,7 +23,7 @@ $ cp config/autopilot-session.example.yaml config/autopilot-session.yaml
 $ node mods/autopilot/dist/session/main.js --config config/autopilot-session.yaml
 ```
 
-See [docs/autopilot.md](../../docs/autopilot.md) for the engine, pacing and the assistant
+See [docs/autopilot.md](../../docs/autopilot.md) for the engine, pacing and the agent
 schema.
 
 ## Example
@@ -64,7 +64,7 @@ The configuration file has two major sections: `conversationSettings` and `langu
 To run the Autopilot, you can use the following command:
 
 ```sh-session
-$ ASSISTANTS=$(pwd)/assistants.json OPEN_API_KEY=your-key autopilot
+$ AGENT=$(pwd)/agents.json OPEN_API_KEY=your-key autopilot
 ```
 
 You will see a message like this:
@@ -187,8 +187,8 @@ For example:
 
 Autopilot reads `autopilot-session.yaml`, found via `--config <path>`, the
 `AUTOPILOT_SESSION_CONFIG` variable or `./config/autopilot-session.yaml`. It names the signing
-secret the gateway uses for Autopilot apps, where apps and assistants come from, the registry
+secret AppBridge uses for Autopilot apps, where apps and agents come from, the registry
 catalog, platform keys by credential ref and the knowledge settings. See
 `config/autopilot-session.example.yaml` for every key.
 
-Audio filters are part of the app's configuration (the gateway runs them), not the assistant's.
+Audio filters are part of the app's configuration (AppBridge runs them), not the agent's.

package/dist/knowledge/MeteredEmbeddings.d.ts

@@ -0,0 +1,47 @@
+/**
+ * Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
+ * http://github.com/fonoster/fonoster
+ *
+ * This file is part of Fonoster
+ *
+ * Licensed under the MIT License (the "License");
+ * you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ *    https://opensource.org/licenses/MIT
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+import { Usage } from "@fonoster/common";
+/** Where embedding usage goes (the session's usage sink). */
+export type EmbeddingUsageSink = {
+    usage(lines: Usage.UsageLine[]): void;
+    missing(catalogId: string): void;
+};
+/**
+ * OpenAI embeddings that report their token usage (add-usage-metering): the knowledge base's
+ * embeddings, called directly so the vendor's `usage.prompt_tokens` is not lost. Same defaults
+ * as LangChain's OpenAIEmbeddings (model text-embedding-ada-002, 512 texts per request); it
+ * implements the embedDocuments/embedQuery pair LangChain's vector stores use.
+ */
+export declare class MeteredEmbeddings {
+    private readonly options;
+    private readonly model;
+    private readonly baseUrl;
+    constructor(options: {
+        apiKey: string;
+        keySource: Usage.KeySource;
+        sink?: EmbeddingUsageSink;
+        model?: string;
+        baseUrl?: string;
+        batchSize?: number;
+    });
+    get catalogId(): string;
+    embedDocuments(texts: string[]): Promise<number[][]>;
+    embedQuery(text: string): Promise<number[]>;
+    private embed;
+}

package/dist/knowledge/MeteredEmbeddings.js

@@ -0,0 +1,58 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.MeteredEmbeddings = void 0;
+/**
+ * OpenAI embeddings that report their token usage (add-usage-metering): the knowledge base's
+ * embeddings, called directly so the vendor's `usage.prompt_tokens` is not lost. Same defaults
+ * as LangChain's OpenAIEmbeddings (model text-embedding-ada-002, 512 texts per request); it
+ * implements the embedDocuments/embedQuery pair LangChain's vector stores use.
+ */
+class MeteredEmbeddings {
+    constructor(options) {
+        this.options = options;
+        this.model = options.model ?? "text-embedding-ada-002";
+        this.baseUrl = (options.baseUrl ?? "https://api.openai.com").replace(/\/$/, "");
+    }
+    get catalogId() {
+        return `openai/${this.model}`;
+    }
+    async embedDocuments(texts) {
+        const size = this.options.batchSize ?? 512;
+        const batches = Array.from({ length: Math.ceil(texts.length / size) }, (_, i) => texts.slice(i * size, (i + 1) * size));
+        const vectors = await Promise.all(batches.map((b) => this.embed(b)));
+        return vectors.flat();
+    }
+    async embedQuery(text) {
+        return (await this.embed([text]))[0];
+    }
+    async embed(input) {
+        const res = await fetch(`${this.baseUrl}/v1/embeddings`, {
+            method: "POST",
+            headers: {
+                "content-type": "application/json",
+                authorization: `Bearer ${this.options.apiKey}`
+            },
+            body: JSON.stringify({ model: this.model, input })
+        });
+        if (!res.ok)
+            throw new Error(`embeddings answered HTTP ${res.status}`);
+        const body = (await res.json());
+        const tokens = body.usage?.prompt_tokens;
+        if (tokens === undefined)
+            this.options.sink?.missing(this.catalogId);
+        else
+            this.options.sink?.usage([
+                {
+                    meter: "embedding.tokens",
+                    quantity: tokens,
+                    provider: "openai",
+                    model: this.catalogId,
+                    keySource: this.options.keySource
+                }
+            ]);
+        return [...body.data]
+            .sort((a, b) => a.index - b.index)
+            .map((d) => d.embedding);
+    }
+}
+exports.MeteredEmbeddings = MeteredEmbeddings;

package/dist/models/createLanguageModel.d.ts

@@ -16,10 +16,10 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-import { EvalAssistant, KnowledgeBase, TelephonyContext, Voice } from "..";
+import { EvalAgent, KnowledgeBase, TelephonyContext, Voice } from "..";
 declare function createLanguageModel(params: {
     voice: Voice;
-    assistantConfig: EvalAssistant;
+    agentConfig: EvalAgent;
     knowledgeBase: KnowledgeBase;
     telephonyContext: TelephonyContext;
 }): import("./AbstractLanguageModel").AbstractLanguageModel;

package/dist/models/createLanguageModel.js

@@ -21,8 +21,8 @@ exports.createLanguageModel = createLanguageModel;
  */
 const __1 = require("..");
 function createLanguageModel(params) {
-    const { voice, assistantConfig, knowledgeBase, telephonyContext } = params;
-    const { languageModel, conversationSettings } = assistantConfig;
+    const { voice, agentConfig, knowledgeBase, telephonyContext } = params;
+    const { languageModel, conversationSettings } = agentConfig;
     // The transfer tool is only added if the transfer options exist
     const tools = languageModel.tools.concat(conversationSettings.transferOptions
         ? [__1.hangupToolDefinition, __1.transferToolDefinition]

package/dist/models/evaluations/createEvalEffectiveConfig.d.ts

@@ -16,13 +16,13 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-import { SessionAssistant } from "@fonoster/common";
-import { EvalAssistant } from "../types";
+import { SessionAgent } from "@fonoster/common";
+import { EvalAgent } from "../types";
 /**
- * The session assistant as the evaluations run it: the catalog id split into provider and model,
- * with the keys the caller resolved (never stored in the assistant).
+ * The session agent as the evaluations run it: the catalog id split into provider and model,
+ * with the keys the caller resolved (never stored in the agent).
  */
-export declare function createEvalEffectiveConfig(assistant: SessionAssistant, keys: {
+export declare function createEvalEffectiveConfig(agent: SessionAgent, keys: {
     apiKey?: string;
     evaluationApiKey?: string;
-}): EvalAssistant;
+}): EvalAgent;

package/dist/models/evaluations/createEvalEffectiveConfig.js

@@ -28,20 +28,20 @@ const LLM_PROVIDERS = [
     "anthropic"
 ];
 /**
- * The session assistant as the evaluations run it: the catalog id split into provider and model,
- * with the keys the caller resolved (never stored in the assistant).
+ * The session agent as the evaluations run it: the catalog id split into provider and model,
+ * with the keys the caller resolved (never stored in the agent).
  */
-function createEvalEffectiveConfig(assistant, keys) {
-    const [provider, ...rest] = assistant.languageModel.model.split("/");
+function createEvalEffectiveConfig(agent, keys) {
+    const [provider, ...rest] = agent.languageModel.model.split("/");
     if (!LLM_PROVIDERS.includes(provider))
         throw new Error(`evaluations do not support ${provider} models`);
-    const testCases = common_1.testCasesSchema.parse(assistant.testCases);
-    const { provider: _p, apiKey: _k, baseUrl: _b, ...languageModel } = assistant.languageModel;
+    const testCases = common_1.testCasesSchema.parse(agent.testCases);
+    const { provider: _p, apiKey: _k, baseUrl: _b, ...languageModel } = agent.languageModel;
     void _p;
     void _k;
     void _b;
     return {
-        conversationSettings: assistant.conversationSettings,
+        conversationSettings: agent.conversationSettings,
         languageModel: {
             ...languageModel,
             provider: provider,

package/dist/models/evaluations/createEvaluateIntelligence.d.ts

@@ -1,3 +1,4 @@
+import { Usage } from "@fonoster/common";
 import { EvaluateIntelligenceRequest } from "./types";
 type ServerStreamCall = {
     request: EvaluateIntelligenceRequest;
@@ -8,10 +9,20 @@ type ServerStreamCall = {
  * A model provider's key for a workspace: the workspace's own vendor key, else the platform key
  * by the provider's credentialRef (the control plane decides; integrations.json is gone).
  */
-export type EvalKeySource = (providerId: string, workspaceRef: string) => Promise<string | undefined>;
+export type EvalKeySource = (providerId: string, workspaceRef: string) => Promise<EvalKey | string | undefined>;
+/** A key and whose it is; a bare string is the platform's. */
+export type EvalKey = {
+    key: string;
+    source: Usage.KeySource;
+};
+/**
+ * Where an eval's usage goes (add-usage-metering D5): the workspace, the lines, and a scope that
+ * is unique per request within the run, for the lines' ids.
+ */
+export type EvalUsageReporter = (workspaceRef: string, lines: Usage.UsageLine[], scope: string) => void;
 /**
- * Runs an assistant's test cases (rework-autopilot 8.2: the assistant is a session assistant; its
- * catalog id picks the provider, and keys come from `keys`, never from the assistant).
+ * Runs an agent's test cases (rework-autopilot 8.2: the agent is a session agent; its
+ * catalog id picks the provider, and keys come from `keys`, never from the agent).
  */
-declare function createEvaluateIntelligence(keys: EvalKeySource): (call: ServerStreamCall) => Promise<void>;
+declare function createEvaluateIntelligence(keys: EvalKeySource, reportUsage?: EvalUsageReporter): (call: ServerStreamCall) => Promise<void>;
 export { createEvaluateIntelligence };

package/dist/models/evaluations/createEvaluateIntelligence.js

@@ -19,33 +19,45 @@ exports.createEvaluateIntelligence = createEvaluateIntelligence;
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
+const crypto_1 = require("crypto");
 const common_1 = require("@fonoster/common");
 const logger_1 = require("@fonoster/logger");
 const pb_util_1 = require("pb-util");
+const meteredThink_1 = require("../../session/meteredThink");
 const createEvalEffectiveConfig_1 = require("./createEvalEffectiveConfig");
 const runEval_1 = require("./runEval");
 const session_1 = require("./session");
 const stepReportToEventPayload_1 = require("./stepReportToEventPayload");
 const logger = (0, logger_1.getLogger)({ service: "apiserver", filePath: __filename });
+const keyOf = (k) => typeof k === "string" ? { key: k, source: "platform" } : k;
 /**
- * Runs an assistant's test cases (rework-autopilot 8.2: the assistant is a session assistant; its
- * catalog id picks the provider, and keys come from `keys`, never from the assistant).
+ * Runs an agent's test cases (rework-autopilot 8.2: the agent is a session agent; its
+ * catalog id picks the provider, and keys come from `keys`, never from the agent).
  */
-function createEvaluateIntelligence(keys) {
+function createEvaluateIntelligence(keys, reportUsage) {
     const evaluateIntelligence = async (call) => {
         const { request } = call;
         const { intelligence } = request;
         const accessKeyId = (0, common_1.getAccessKeyIdFromCall)(call);
         try {
             const config = pb_util_1.struct.decode(intelligence.config);
-            const assistant = common_1.sessionAssistantSchema.parse(config);
-            const provider = assistant.languageModel.model.split("/")[0];
+            const agent = common_1.sessionAgentSchema.parse(config);
+            const provider = agent.languageModel.model.split("/")[0];
             logger.verbose("call to evaluateIntelligence", {
                 accessKeyId,
-                model: assistant.languageModel.model
+                model: agent.languageModel.model
             });
-            const apiKey = await keys(provider, accessKeyId);
-            const evaluationApiKey = await keys("openai", accessKeyId);
+            const agentKey = keyOf(await keys(provider, accessKeyId));
+            const judgeKey = keyOf(await keys("openai", accessKeyId));
+            const apiKey = agentKey?.key;
+            const evaluationApiKey = judgeKey?.key;
+            // Eval tokens are metered with origin eval, one scope per request of this run.
+            const runId = (0, crypto_1.randomUUID)();
+            let request = 0;
+            const sink = {
+                usage: (lines) => reportUsage?.(accessKeyId, lines, `eval|${runId}|${request++}`),
+                missing: (catalogId) => logger.warn(`eval: ${catalogId} reported no usage`, { accessKeyId })
+            };
             if (!apiKey && provider !== "ollama")
                 throw new Error(`no ${provider} key for this workspace or platform`);
             if (!evaluationApiKey)
@@ -60,12 +72,21 @@ function createEvaluateIntelligence(keys) {
             };
             // The session engine (streaming replies, structured tool messages) for OpenAI-compatible
             // providers; the LangChain path stays for the rest until their adapters land (EV-B1).
-            const think = (0, session_1.sessionThinkFor)(assistant, apiKey);
-            if (think) {
-                await (0, session_1.runSessionEval)({ assistant, think, evaluationApiKey }, callbacks);
+            const sessionThink = (0, session_1.sessionThinkFor)(agent, apiKey);
+            if (sessionThink) {
+                const think = (0, meteredThink_1.meteredThink)(sessionThink, {
+                    catalogId: agent.languageModel.model,
+                    keySource: agentKey?.source ?? "platform"
+                }, sink);
+                await (0, session_1.runSessionEval)({
+                    agent,
+                    think,
+                    evaluationApiKey,
+                    judgeUsage: { sink, keySource: judgeKey?.source ?? "platform" }
+                }, callbacks);
                 return;
             }
-            const effectiveConfig = (0, createEvalEffectiveConfig_1.createEvalEffectiveConfig)(assistant, {
+            const effectiveConfig = (0, createEvalEffectiveConfig_1.createEvalEffectiveConfig)(agent, {
                 apiKey,
                 evaluationApiKey
             });

package/dist/models/evaluations/createTestTextSimilarity.d.ts

@@ -1,6 +1,12 @@
+import { MeteredModel, UsageSink } from "../../session/meteredThink";
 export declare function createTestTextSimilarity(evalsLanguageModel: {
     provider: any;
     model: string;
     baseUrl?: string;
     apiKey?: string;
-}, systemPrompt: string): (text1: string, text2: string) => Promise<boolean>;
+}, systemPrompt: string, 
+/** Meters the judge's tokens (add-usage-metering); the judge is always OpenAI. */
+usage?: {
+    sink: UsageSink;
+    keySource: MeteredModel["keySource"];
+}): (text1: string, text2: string) => Promise<boolean>;

package/dist/models/evaluations/createTestTextSimilarity.js

@@ -21,7 +21,10 @@ exports.createTestTextSimilarity = createTestTextSimilarity;
  */
 const messages_1 = require("@langchain/core/messages");
 const openai_1 = require("@langchain/openai");
-function createTestTextSimilarity(evalsLanguageModel, systemPrompt) {
+const meteredThink_1 = require("../../session/meteredThink");
+function createTestTextSimilarity(evalsLanguageModel, systemPrompt, 
+/** Meters the judge's tokens (add-usage-metering); the judge is always OpenAI. */
+usage) {
     if (!evalsLanguageModel.apiKey) {
         throw new Error("API key is required for text similarity evaluation.");
     }
@@ -37,6 +40,21 @@ function createTestTextSimilarity(evalsLanguageModel, systemPrompt) {
             new messages_1.HumanMessage(`Text 1: ${text1}\nText 2: ${text2}`)
         ];
         const response = await llm.invoke(messages);
+        if (usage) {
+            const model = {
+                catalogId: `openai/${evalsLanguageModel.model}`,
+                keySource: usage.keySource
+            };
+            const m = response.usage_metadata;
+            if (m)
+                usage.sink.usage((0, meteredThink_1.usageLines)(model, {
+                    inputTokens: m.input_tokens,
+                    outputTokens: m.output_tokens,
+                    cachedTokens: m.input_token_details?.cache_read ?? 0
+                }));
+            else
+                usage.sink.missing(model.catalogId);
+        }
         const reply = response.content?.toString().trim().toLowerCase();
         return reply === "true" || reply === "yes";
     };

package/dist/models/evaluations/evalTestCases.d.ts

@@ -17,9 +17,9 @@
  * limitations under the License.
  */
 import { ScenarioEvaluationReport } from "@fonoster/types";
-import { EvalAssistant } from "../types";
+import { EvalAgent } from "../types";
 export declare function evalTestCases(autopilotApplication: {
     intelligence: {
-        config: EvalAssistant;
+        config: EvalAgent;
     };
 }): Promise<ScenarioEvaluationReport[]>;

package/dist/models/evaluations/evalTestCases.js

@@ -7,7 +7,7 @@ const evaluateScenario_1 = require("./evaluateScenario");
 const textSimilaryPrompt_1 = require("./textSimilaryPrompt");
 async function evalTestCases(autopilotApplication) {
     const { testCases } = autopilotApplication.intelligence.config;
-    const { config: assistantConfig } = autopilotApplication.intelligence;
+    const { config: agentConfig } = autopilotApplication.intelligence;
     const voice = {
         say: async (_) => { }
     };
@@ -17,7 +17,7 @@ async function evalTestCases(autopilotApplication) {
         const scenario = scenarios[i];
         const languageModel = (0, createLanguageModel_1.createLanguageModel)({
             voice,
-            assistantConfig: autopilotApplication.intelligence.config,
+            agentConfig: autopilotApplication.intelligence.config,
             knowledgeBase: {
                 load: async () => { },
                 queryKnowledgeBase: async (query) => query
@@ -25,12 +25,12 @@ async function evalTestCases(autopilotApplication) {
             telephonyContext: scenario.telephonyContext
         });
         const testTextSimilarity = (0, createTestTextSimilarity_1.createTestTextSimilarity)({
-            provider: assistantConfig.testCases?.evalsLanguageModel?.provider,
-            model: assistantConfig.testCases?.evalsLanguageModel?.model ?? "",
-            apiKey: assistantConfig.testCases?.evalsLanguageModel?.apiKey
-        }, assistantConfig.testCases?.evalsSystemPrompt || textSimilaryPrompt_1.textSimilaryPrompt);
+            provider: agentConfig.testCases?.evalsLanguageModel?.provider,
+            model: agentConfig.testCases?.evalsLanguageModel?.model ?? "",
+            apiKey: agentConfig.testCases?.evalsLanguageModel?.apiKey
+        }, agentConfig.testCases?.evalsSystemPrompt || textSimilaryPrompt_1.textSimilaryPrompt);
         const evaluationReport = await (0, evaluateScenario_1.evaluateScenario)({
-            assistantConfig,
+            agentConfig,
             scenario,
             languageModel,
             testTextSimilarity

package/dist/models/evaluations/evaluateScenario.js

@@ -3,7 +3,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.evaluateScenario = evaluateScenario;
 const evaluateStep_1 = require("./evaluateStep");
 async function evaluateScenario(config, onStepResult) {
-    const { scenario, languageModel, testTextSimilarity, assistantConfig } = config;
+    const { scenario, languageModel, testTextSimilarity, agentConfig } = config;
     const results = [];
     const steps = scenario.conversation;
     for (let i = 0; i < steps.length; i++) {
@@ -12,7 +12,7 @@ async function evaluateScenario(config, onStepResult) {
             step,
             languageModel,
             testTextSimilarity,
-            assistantConfig
+            agentConfig
         });
         results.push(stepResult);
         if (onStepResult)

package/dist/models/evaluations/evaluateStep.d.ts

@@ -18,4 +18,4 @@
  */
 import { StepEvaluationReport } from "@fonoster/types";
 import { EvaluateStepParams } from "./types";
-export declare function evaluateStep({ step, languageModel, testTextSimilarity, assistantConfig }: EvaluateStepParams): Promise<StepEvaluationReport>;
+export declare function evaluateStep({ step, languageModel, testTextSimilarity, agentConfig }: EvaluateStepParams): Promise<StepEvaluationReport>;

package/dist/models/evaluations/evaluateStep.js

@@ -4,7 +4,7 @@ exports.evaluateStep = evaluateStep;
 const evaluateTextResponse_1 = require("./evaluateTextResponse");
 const evaluateToolCalls_1 = require("./evaluateToolCalls");
 const extractAiResponse_1 = require("./extractAiResponse");
-async function evaluateStep({ step, languageModel, testTextSimilarity, assistantConfig }) {
+async function evaluateStep({ step, languageModel, testTextSimilarity, agentConfig }) {
     const stepResult = {
         humanInput: step.userInput,
         expectedResponse: step.expected.text.response,
@@ -14,7 +14,7 @@ async function evaluateStep({ step, languageModel, testTextSimilarity, assistant
     };
     try {
         const response = await languageModel.invoke(step.userInput);
-        stepResult.aiResponse = (0, extractAiResponse_1.extractAiResponse)(response, assistantConfig);
+        stepResult.aiResponse = (0, extractAiResponse_1.extractAiResponse)(response, agentConfig);
         const textEvaluation = await (0, evaluateTextResponse_1.evaluateTextResponse)({
             type: step.expected.text.type,
             response: step.expected.text.response

package/dist/models/evaluations/extractAiResponse.d.ts

@@ -16,9 +16,9 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-import { EvalAssistant, InvocationResult } from "../types";
+import { EvalAgent, InvocationResult } from "../types";
 /**
  * Maps language model invocation result to the text that should be compared
  * in eval (hangup/transfer use config messages).
  */
-export declare function extractAiResponse(response: InvocationResult, assistantConfig: EvalAssistant): string;
+export declare function extractAiResponse(response: InvocationResult, agentConfig: EvalAgent): string;

package/dist/models/evaluations/extractAiResponse.js

@@ -5,14 +5,14 @@ exports.extractAiResponse = extractAiResponse;
  * Maps language model invocation result to the text that should be compared
  * in eval (hangup/transfer use config messages).
  */
-function extractAiResponse(response, assistantConfig) {
+function extractAiResponse(response, agentConfig) {
     if (response.toolCalls && response.toolCalls.length > 0) {
         const topTool = response.toolCalls[0];
         if (topTool.name === "hangup") {
-            return assistantConfig.conversationSettings?.goodbyeMessage ?? "";
+            return agentConfig.conversationSettings?.goodbyeMessage ?? "";
         }
         if (topTool.name === "transfer") {
-            return (assistantConfig.conversationSettings?.transferOptions?.message ?? "");
+            return agentConfig.conversationSettings?.transferOptions?.message ?? "";
         }
     }
     return response.content ?? "";

package/dist/models/evaluations/runEval.d.ts

@@ -1,8 +1,8 @@
-import { EvalAssistant } from "../types";
+import { EvalAgent } from "../types";
 import { RunEvalCallbacks } from "./types";
 export type RunEvalConfig = {
     intelligence: {
-        config: EvalAssistant;
+        config: EvalAgent;
     };
 };
 export declare function runEval(autopilotApplication: RunEvalConfig, callbacks: RunEvalCallbacks): Promise<void>;

package/dist/models/evaluations/runEval.js

@@ -7,7 +7,7 @@ const evaluateScenario_1 = require("./evaluateScenario");
 const textSimilaryPrompt_1 = require("./textSimilaryPrompt");
 async function runEval(autopilotApplication, callbacks) {
     const { testCases } = autopilotApplication.intelligence.config;
-    const assistantConfig = autopilotApplication.intelligence.config;
+    const agentConfig = autopilotApplication.intelligence.config;
     const voice = {
         say: async (_) => { }
     };
@@ -15,7 +15,7 @@ async function runEval(autopilotApplication, callbacks) {
     for (const scenario of scenarios) {
         const languageModel = (0, createLanguageModel_1.createLanguageModel)({
             voice,
-            assistantConfig,
+            agentConfig,
             knowledgeBase: {
                 load: async () => { },
                 queryKnowledgeBase: async (query) => query
@@ -23,12 +23,12 @@ async function runEval(autopilotApplication, callbacks) {
             telephonyContext: scenario.telephonyContext
         });
         const testTextSimilarity = (0, createTestTextSimilarity_1.createTestTextSimilarity)({
-            provider: assistantConfig.testCases?.evalsLanguageModel?.provider,
-            model: assistantConfig.testCases?.evalsLanguageModel?.model ?? "",
-            apiKey: assistantConfig.testCases?.evalsLanguageModel?.apiKey
-        }, assistantConfig.testCases?.evalsSystemPrompt ?? textSimilaryPrompt_1.textSimilaryPrompt);
+            provider: agentConfig.testCases?.evalsLanguageModel?.provider,
+            model: agentConfig.testCases?.evalsLanguageModel?.model ?? "",
+            apiKey: agentConfig.testCases?.evalsLanguageModel?.apiKey
+        }, agentConfig.testCases?.evalsSystemPrompt ?? textSimilaryPrompt_1.textSimilaryPrompt);
         const report = await (0, evaluateScenario_1.evaluateScenario)({
-            assistantConfig,
+            agentConfig,
             scenario,
             languageModel,
             testTextSimilarity

package/dist/models/evaluations/session/EvalSession.d.ts

@@ -26,7 +26,7 @@ export type EvalTelephony = {
 };
 /**
  * An in-memory session for conversation evals (rework-autopilot 1.4): the session engine runs on
- * it exactly as on a call, but there is no gateway, socket or audio. Turns go in as `turn.ended`;
+ * it exactly as on a call, but there is no AppBridge, socket or audio. Turns go in as `turn.ended`;
  * every reply is "heard" in full (no barge-in in an eval).
  */
 export declare class EvalSession {

package/dist/models/evaluations/session/EvalSession.js

@@ -23,7 +23,7 @@ const events_1 = require("events");
 const voice_1 = require("@fonoster/voice");
 /**
  * An in-memory session for conversation evals (rework-autopilot 1.4): the session engine runs on
- * it exactly as on a call, but there is no gateway, socket or audio. Turns go in as `turn.ended`;
+ * it exactly as on a call, but there is no AppBridge, socket or audio. Turns go in as `turn.ended`;
  * every reply is "heard" in full (no barge-in in an eval).
  */
 class EvalSession {

package/dist/models/evaluations/session/evaluateSessionScenario.d.ts

@@ -17,7 +17,7 @@
  * limitations under the License.
  */
 import { ScenarioEvaluationReport, StepEvaluationReport } from "@fonoster/types";
-import { SessionAssistant } from "../../../session/config";
+import { SessionAgent } from "../../../session/config";
 import { ChatMessage, Think } from "../../../session/think";
 import { ToolPort } from "../../../session/tools";
 import { EvalScenario } from "../types";
@@ -36,12 +36,12 @@ export type SessionScenarioReport = ScenarioEvaluationReport & {
     steps: SessionStepReport[];
 };
 export type SessionEvalOptions = {
-    assistant: SessionAssistant;
+    agent: SessionAgent;
     /** The language model the engine streams from (a registry adapter, or the fake LLM). */
     think: Think;
     testTextSimilarity: (text1: string, text2: string) => Promise<boolean>;
     /**
-     * Runs the assistant's own tools. Defaults to `evalToolPort`: evals never call tool endpoints
+     * Runs the agent's own tools. Defaults to `evalToolPort`: evals never call tool endpoints
      * (an eval must not book an appointment), every call succeeds with `{ status: "ok" }`.
      */
     tools?: ToolPort;

package/dist/models/evaluations/session/evaluateSessionScenario.js

@@ -41,7 +41,7 @@ function parseArgs(call) {
  * read back from the structured tool messages in the history.
  */
 async function evaluateSessionScenario(options, scenario, onStepResult) {
-    const { assistant, testTextSimilarity } = options;
+    const { agent: agentConfig, testTextSimilarity } = options;
     const stepTimeoutMs = options.stepTimeoutMs ?? 60000;
     const session = new EvalSession_1.EvalSession(scenario.telephonyContext);
     let agent;
@@ -64,7 +64,7 @@ async function evaluateSessionScenario(options, scenario, onStepResult) {
         })()
     };
     const handler = (0, createAutopilot_1.createAutopilotHandler)({
-        assistant: { ...assistant, testCases: undefined },
+        agent: { ...agentConfig, testCases: undefined },
         think,
         tools: options.tools ?? exports.evalToolPort,
         recorder: {
@@ -134,8 +134,8 @@ async function evaluateSessionScenario(options, scenario, onStepResult) {
             // engine says for it (as the legacy evals compared).
             report.aiResponse = builtIn
                 ? builtIn.name === "hangup"
-                    ? assistant.conversationSettings.goodbyeMessage
-                    : (assistant.conversationSettings.transferOptions?.message ?? "")
+                    ? agentConfig.conversationSettings.goodbyeMessage
+                    : (agentConfig.conversationSettings.transferOptions?.message ?? "")
                 : reply
                     .map((s) => s.text)
                     .join("")