@openclaw/voice-call 2026.5.2 → 2026.5.3-beta.1

package/src/response-generator.ts

@@ -1,318 +0,0 @@
-/**
- * Voice call response generator - uses the embedded Pi agent for tool support.
- * Routes voice responses through the same agent infrastructure as messaging.
- */
-
-import crypto from "node:crypto";
-import { applyModelOverrideToSessionEntry } from "openclaw/plugin-sdk/model-session-runtime";
-import { normalizeLowercaseStringOrEmpty } from "openclaw/plugin-sdk/text-runtime";
-import type { SessionEntry } from "../api.js";
-import { resolveVoiceCallSessionKey, type VoiceCallConfig } from "./config.js";
-import type { CoreAgentDeps, CoreConfig } from "./core-bridge.js";
-import { resolveVoiceResponseModel } from "./response-model.js";
-
-export type VoiceResponseParams = {
-  /** Voice call config */
-  voiceConfig: VoiceCallConfig;
-  /** Core OpenClaw config */
-  coreConfig: CoreConfig;
-  /** Injected host agent runtime */
-  agentRuntime: CoreAgentDeps;
-  /** Call ID for session tracking */
-  callId: string;
-  /** Persisted call session key */
-  sessionKey?: string;
-  /** Caller's phone number */
-  from: string;
-  /** Conversation transcript */
-  transcript: Array<{ speaker: "user" | "bot"; text: string }>;
-  /** Latest user message */
-  userMessage: string;
-};
-
-export type VoiceResponseResult = {
-  text: string | null;
-  error?: string;
-};
-
-type VoiceResponsePayload = {
-  text?: string;
-  isError?: boolean;
-  isReasoning?: boolean;
-};
-
-const VOICE_SPOKEN_OUTPUT_CONTRACT = [
-  "Output format requirements:",
-  '- Return only valid JSON in this exact shape: {"spoken":"..."}',
-  "- Do not include markdown, code fences, planning text, or extra keys.",
-  '- Put exactly what should be spoken to the caller into "spoken".',
-  '- If there is nothing to say, return {"spoken":""}.',
-].join("\n");
-
-function normalizeSpokenText(value: string): string | null {
-  const normalized = value.replace(/\s+/g, " ").trim();
-  return normalized.length > 0 ? normalized : null;
-}
-
-function tryParseSpokenJson(text: string): string | null {
-  const candidates: string[] = [];
-  const trimmed = text.trim();
-  if (!trimmed) {
-    return null;
-  }
-  candidates.push(trimmed);
-
-  const fenced = trimmed.match(/^```(?:json)?\s*([\s\S]*?)\s*```$/i);
-  if (fenced?.[1]) {
-    candidates.push(fenced[1]);
-  }
-
-  const firstBrace = trimmed.indexOf("{");
-  const lastBrace = trimmed.lastIndexOf("}");
-  if (firstBrace >= 0 && lastBrace > firstBrace) {
-    candidates.push(trimmed.slice(firstBrace, lastBrace + 1));
-  }
-
-  for (const candidate of candidates) {
-    try {
-      const parsed = JSON.parse(candidate) as { spoken?: unknown };
-      if (typeof parsed?.spoken !== "string") {
-        continue;
-      }
-      return normalizeSpokenText(parsed.spoken) ?? "";
-    } catch {
-      // Continue trying other candidates.
-    }
-  }
-
-  const inlineSpokenMatch = trimmed.match(/"spoken"\s*:\s*"((?:[^"\\]|\\.)*)"/i);
-  if (!inlineSpokenMatch) {
-    return null;
-  }
-
-  try {
-    const decoded = JSON.parse(`"${inlineSpokenMatch[1] ?? ""}"`) as string;
-    return normalizeSpokenText(decoded) ?? "";
-  } catch {
-    return null;
-  }
-}
-
-function isLikelyMetaReasoningParagraph(paragraph: string): boolean {
-  const lower = normalizeLowercaseStringOrEmpty(paragraph);
-  if (!lower) {
-    return false;
-  }
-
-  if (lower.startsWith("thinking process")) {
-    return true;
-  }
-  if (lower.startsWith("reasoning:") || lower.startsWith("analysis:")) {
-    return true;
-  }
-  if (
-    lower.startsWith("the user ") &&
-    (lower.includes("i should") || lower.includes("i need to") || lower.includes("i will"))
-  ) {
-    return true;
-  }
-  if (
-    lower.includes("this is a natural continuation of the conversation") ||
-    lower.includes("keep the conversation flowing")
-  ) {
-    return true;
-  }
-
-  return false;
-}
-
-function sanitizePlainSpokenText(text: string): string | null {
-  const withoutCodeFences = text.replace(/```[\s\S]*?```/g, " ").trim();
-  if (!withoutCodeFences) {
-    return null;
-  }
-
-  const paragraphs = withoutCodeFences
-    .split(/\n\s*\n+/)
-    .map((paragraph) => paragraph.trim())
-    .filter(Boolean);
-
-  while (paragraphs.length > 1 && isLikelyMetaReasoningParagraph(paragraphs[0])) {
-    paragraphs.shift();
-  }
-
-  return normalizeSpokenText(paragraphs.join(" "));
-}
-
-function extractSpokenTextFromPayloads(payloads: VoiceResponsePayload[]): string | null {
-  const spokenSegments: string[] = [];
-
-  for (const payload of payloads) {
-    if (payload.isError || payload.isReasoning) {
-      continue;
-    }
-
-    const rawText = payload.text?.trim() ?? "";
-    if (!rawText) {
-      continue;
-    }
-
-    const structured = tryParseSpokenJson(rawText);
-    if (structured !== null) {
-      if (structured.length > 0) {
-        spokenSegments.push(structured);
-      }
-      continue;
-    }
-
-    const plain = sanitizePlainSpokenText(rawText);
-    if (plain) {
-      spokenSegments.push(plain);
-    }
-  }
-
-  return spokenSegments.length > 0 ? spokenSegments.join(" ").trim() : null;
-}
-
-function resolveVoiceSandboxSessionKey(agentId: string, sessionKey: string): string {
-  const trimmed = sessionKey.trim();
-  if (trimmed.toLowerCase().startsWith("agent:")) {
-    return trimmed;
-  }
-  return `agent:${agentId}:${trimmed}`;
-}
-
-/**
- * Generate a voice response using the embedded Pi agent with full tool support.
- * Uses the same agent infrastructure as messaging for consistent behavior.
- */
-export async function generateVoiceResponse(
-  params: VoiceResponseParams,
-): Promise<VoiceResponseResult> {
-  const {
-    voiceConfig,
-    callId,
-    sessionKey,
-    from,
-    transcript,
-    userMessage,
-    coreConfig,
-    agentRuntime,
-  } = params;
-
-  if (!coreConfig) {
-    return { text: null, error: "Core config unavailable for voice response" };
-  }
-  const cfg = coreConfig;
-
-  const resolvedSessionKey = resolveVoiceCallSessionKey({
-    config: voiceConfig,
-    callId,
-    phone: from,
-    explicitSessionKey: sessionKey,
-  });
-  const agentId = voiceConfig.agentId ?? "main";
-
-  // Resolve paths
-  const storePath = agentRuntime.session.resolveStorePath(cfg.session?.store, { agentId });
-  const agentDir = agentRuntime.resolveAgentDir(cfg, agentId);
-  const workspaceDir = agentRuntime.resolveAgentWorkspaceDir(cfg, agentId);
-
-  // Ensure workspace exists
-  await agentRuntime.ensureAgentWorkspace({ dir: workspaceDir });
-
-  // Load or create session entry
-  const sessionStore = agentRuntime.session.loadSessionStore(storePath);
-  const now = Date.now();
-  const existingSessionEntry = sessionStore[resolvedSessionKey] as SessionEntry | undefined;
-
-  // Resolve model from config
-  const { provider, model } = resolveVoiceResponseModel({ voiceConfig, agentRuntime });
-
-  let sessionEntry = existingSessionEntry;
-  if (!sessionEntry?.sessionId || voiceConfig.responseModel) {
-    sessionEntry = await agentRuntime.session.updateSessionStore(storePath, (store) => {
-      let entry = store[resolvedSessionKey] as SessionEntry | undefined;
-      if (!entry?.sessionId) {
-        entry = {
-          ...entry,
-          sessionId: crypto.randomUUID(),
-          updatedAt: now,
-        };
-        store[resolvedSessionKey] = entry;
-      }
-      if (voiceConfig.responseModel) {
-        applyModelOverrideToSessionEntry({
-          entry,
-          selection: { provider, model },
-          selectionSource: "auto",
-        });
-      }
-      return entry;
-    });
-  }
-  const sessionId = sessionEntry.sessionId;
-
-  const sessionFile = agentRuntime.session.resolveSessionFilePath(sessionId, sessionEntry, {
-    agentId,
-  });
-
-  // Resolve thinking level
-  const thinkLevel = agentRuntime.resolveThinkingDefault({ cfg, provider, model });
-
-  // Resolve agent identity for personalized prompt
-  const identity = agentRuntime.resolveAgentIdentity(cfg, agentId);
-  const agentName = identity?.name?.trim() || "assistant";
-
-  // Build system prompt with conversation history
-  const basePrompt =
-    voiceConfig.responseSystemPrompt ??
-    `You are ${agentName}, a helpful voice assistant on a phone call. Keep responses brief and conversational (1-2 sentences max). Be natural and friendly. The caller's phone number is ${from}. You have access to tools - use them when helpful.`;
-
-  let extraSystemPrompt = basePrompt;
-  if (transcript.length > 0) {
-    const history = transcript
-      .map((entry) => `${entry.speaker === "bot" ? "You" : "Caller"}: ${entry.text}`)
-      .join("\n");
-    extraSystemPrompt = `${basePrompt}\n\nConversation so far:\n${history}`;
-  }
-  extraSystemPrompt = `${extraSystemPrompt}\n\n${VOICE_SPOKEN_OUTPUT_CONTRACT}`;
-
-  // Resolve timeout
-  const timeoutMs = voiceConfig.responseTimeoutMs ?? agentRuntime.resolveAgentTimeoutMs({ cfg });
-  const runId = `voice:${callId}:${Date.now()}`;
-
-  try {
-    const result = await agentRuntime.runEmbeddedPiAgent({
-      sessionId,
-      sessionKey: resolvedSessionKey,
-      sandboxSessionKey: resolveVoiceSandboxSessionKey(agentId, resolvedSessionKey),
-      agentId,
-      messageProvider: "voice",
-      sessionFile,
-      workspaceDir,
-      config: cfg,
-      prompt: userMessage,
-      provider,
-      model,
-      thinkLevel,
-      verboseLevel: "off",
-      timeoutMs,
-      runId,
-      lane: "voice",
-      extraSystemPrompt,
-      agentDir,
-    });
-
-    const text = extractSpokenTextFromPayloads((result.payloads ?? []) as VoiceResponsePayload[]);
-
-    if (!text && result.meta?.aborted) {
-      return { text: null, error: "Response generation was aborted" };
-    }
-
-    return { text };
-  } catch (err) {
-    console.error(`[voice-call] Response generation failed:`, err);
-    return { text: null, error: String(err) };
-  }
-}

package/src/response-model.test.ts

@@ -1,71 +0,0 @@
-import { describe, expect, it } from "vitest";
-import { VoiceCallConfigSchema } from "./config.js";
-import type { CoreAgentDeps } from "./core-bridge.js";
-import { resolveVoiceResponseModel } from "./response-model.js";
-
-const agentRuntime = {
-  defaults: {
-    provider: "together",
-    model: "Qwen/Qwen2.5-7B-Instruct-Turbo",
-  },
-} as unknown as CoreAgentDeps;
-
-describe("resolveVoiceResponseModel", () => {
-  it("falls back to the runtime default model", () => {
-    expect(
-      resolveVoiceResponseModel({
-        voiceConfig: VoiceCallConfigSchema.parse({}),
-        agentRuntime,
-      }),
-    ).toEqual({
-      modelRef: "together/Qwen/Qwen2.5-7B-Instruct-Turbo",
-      provider: "together",
-      model: "Qwen/Qwen2.5-7B-Instruct-Turbo",
-    });
-  });
-
-  it("uses an explicit provider/model ref", () => {
-    expect(
-      resolveVoiceResponseModel({
-        voiceConfig: VoiceCallConfigSchema.parse({
-          responseModel: "openai/gpt-5.4-mini",
-        }),
-        agentRuntime,
-      }),
-    ).toEqual({
-      modelRef: "openai/gpt-5.4-mini",
-      provider: "openai",
-      model: "gpt-5.4-mini",
-    });
-  });
-
-  it("uses the runtime default provider for bare model overrides", () => {
-    expect(
-      resolveVoiceResponseModel({
-        voiceConfig: VoiceCallConfigSchema.parse({
-          responseModel: "meta-llama/Llama-4-Scout-17B-16E-Instruct",
-        }),
-        agentRuntime,
-      }),
-    ).toEqual({
-      modelRef: "meta-llama/Llama-4-Scout-17B-16E-Instruct",
-      provider: "meta-llama",
-      model: "Llama-4-Scout-17B-16E-Instruct",
-    });
-  });
-
-  it("keeps legacy single-segment overrides on the runtime default provider", () => {
-    expect(
-      resolveVoiceResponseModel({
-        voiceConfig: VoiceCallConfigSchema.parse({
-          responseModel: "gpt-5.4-mini",
-        }),
-        agentRuntime,
-      }),
-    ).toEqual({
-      modelRef: "gpt-5.4-mini",
-      provider: "together",
-      model: "gpt-5.4-mini",
-    });
-  });
-});

package/src/response-model.ts

@@ -1,23 +0,0 @@
-import type { VoiceCallConfig } from "./config.js";
-import type { CoreAgentDeps } from "./core-bridge.js";
-
-export function resolveVoiceResponseModel(params: {
-  voiceConfig: VoiceCallConfig;
-  agentRuntime: CoreAgentDeps;
-}): {
-  modelRef: string;
-  provider: string;
-  model: string;
-} {
-  const modelRef =
-    params.voiceConfig.responseModel ??
-    `${params.agentRuntime.defaults.provider}/${params.agentRuntime.defaults.model}`;
-  const slashIndex = modelRef.indexOf("/");
-
-  return {
-    modelRef,
-    provider:
-      slashIndex === -1 ? params.agentRuntime.defaults.provider : modelRef.slice(0, slashIndex),
-    model: slashIndex === -1 ? modelRef : modelRef.slice(slashIndex + 1),
-  };
-}