voicecc 1.1.35 → 1.2.0

package/dashboard/server.ts

@@ -3,6 +3,7 @@
  *
  * Thin wiring file that creates the Hono app and starts listening:
  * - Mount all API route groups under /api/*
+ * - Proxy Twilio media WebSocket upgrades to the Python server
  * - Serve the Vite build output as static files
  * - SPA fallback for client-side routing
  */
@@ -15,6 +16,10 @@ import { readFileSync } from "fs";
 import { access } from "fs/promises";
 import { join } from "path";
 import { homedir } from "os";
+import { WebSocket as WsWebSocket, WebSocketServer } from "ws";
+
+import type { IncomingMessage } from "http";
+import type { Duplex } from "stream";
 
 import { claudeMdRoutes } from "./routes/claude-md.js";
 import { conversationRoutes } from "./routes/conversations.js";
@@ -30,6 +35,7 @@ import { providersRoutes } from "./routes/providers.js";
 import { agentsRoutes } from "./routes/agents.js";
 import { versionRoutes } from "./routes/version.js";
 import { chatRoutes } from "./routes/chat.js";
+import { voiceRoutes } from "./routes/voice.js";
 import { loadDeviceTokens } from "../server/services/device-pairing.js";
 
 // ============================================================================
@@ -39,6 +45,9 @@ import { loadDeviceTokens } from "../server/services/device-pairing.js";
 const PORTS_TO_TRY = [3456, 3457, 3458, 3459, 3460];
 const USER_CLAUDE_MD_PATH = join(homedir(), ".claude", "CLAUDE.md");
 
+/** Base URL for the Python FastAPI server (for WebSocket + HTTP proxy) */
+const VOICE_API_URL = process.env.VOICE_SERVER_URL ?? "http://localhost:7861";
+
 // ============================================================================
 // MAIN HANDLERS
 // ============================================================================
@@ -51,10 +60,19 @@ const USER_CLAUDE_MD_PATH = join(homedir(), ".claude", "CLAUDE.md");
 export function createApp(): Hono {
   const app = new Hono();
 
-  // Dashboard password protection (HTTP Basic Auth)
+  // Dashboard password protection (HTTP Basic Auth).
+  // Exclude /chat and /api/chat/* -- those use device-token auth and opening
+  // them in a new tab can corrupt the browser's cached Basic Auth credentials.
   const dashboardPassword = process.env.DASHBOARD_PASSWORD;
   if (dashboardPassword) {
-    app.use("*", basicAuth({ username: "admin", password: dashboardPassword }));
+    const auth = basicAuth({ username: "admin", password: dashboardPassword });
+    app.use("*", async (c, next) => {
+      const path = c.req.path;
+      if (path === "/chat" || path.startsWith("/api/chat/") || path.startsWith("/api/webrtc/") || path.startsWith("/api/voice/")) {
+        return next();
+      }
+      return auth(c, next);
+    });
   }
 
   // API route groups
@@ -72,6 +90,7 @@ export function createApp(): Hono {
   app.route("/api/agents", agentsRoutes());
   app.route("/api/version", versionRoutes());
   app.route("/api/chat", chatRoutes());
+  app.route("/api/voice", voiceRoutes());
 
   // Status endpoint (user CLAUDE.md conflict check)
   app.get("/api/status", async (c) => {
@@ -122,6 +141,43 @@ export async function startDashboard(): Promise<number> {
           resolve();
         });
         server.on("error", reject);
+
+        // Proxy /media/:token WebSocket upgrades to the Python server
+        const wss = new WebSocketServer({ noServer: true });
+        server.on("upgrade", (req: IncomingMessage, socket: Duplex, head: Buffer) => {
+          const url = req.url ?? "";
+          const match = url.match(/^\/media\/([a-f0-9-]+)(?:\?.*)?$/);
+          if (!match) return; // Not a Twilio media WebSocket -- let it fall through
+
+          const targetWsUrl = VOICE_API_URL.replace(/^http/, "ws") + url;
+          const upstream = new WsWebSocket(targetWsUrl);
+
+          upstream.on("open", () => {
+            wss.handleUpgrade(req, socket, head, (clientWs) => {
+              // Bidirectional message proxy
+              clientWs.on("message", (data) => {
+                if (upstream.readyState === WsWebSocket.OPEN) {
+                  upstream.send(data);
+                }
+              });
+              upstream.on("message", (data) => {
+                if (clientWs.readyState === WsWebSocket.OPEN) {
+                  clientWs.send(data);
+                }
+              });
+
+              clientWs.on("close", () => upstream.close());
+              upstream.on("close", () => clientWs.close());
+              clientWs.on("error", () => upstream.close());
+              upstream.on("error", () => clientWs.close());
+            });
+          });
+
+          upstream.on("error", (err) => {
+            console.error(`[dashboard] Twilio WS proxy error: ${err.message}`);
+            socket.destroy();
+          });
+        });
       });
 
       setDashboardPort(port);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "voicecc",
-  "version": "1.1.35",
+  "version": "1.2.0",
   "description": "Voice Agent Platform running on Claude Code -- create and deploy conversational voice agents with ElevenLabs STT/TTS and VAD",
   "repository": {
     "type": "git",
@@ -35,11 +35,10 @@
     "linux"
   ],
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.2.70",
+    "@anthropic-ai/claude-agent-sdk": "^0.1.8",
     "@anthropic-ai/sdk": "^0.39.0",
     "@hono/node-server": "^1.19.9",
     "archiver": "^7.0.1",
-    "avr-vad": "^1.0.0",
     "cloudflared": "^0.7.1",
     "dotenv": "^16.4.0",
     "hono": "^4.12.0",

package/server/index.ts

@@ -8,7 +8,22 @@
  * - Auto-start Twilio if enabled (requires tunnel)
  */
 
-import { writeFileSync, unlinkSync, mkdirSync } from "node:fs";
+// Global error handlers -- must be registered before any async work to prevent
+// silent crashes from unhandled promise rejections or uncaught exceptions.
+process.on("uncaughtException", (err) => {
+  console.error("[FATAL] Uncaught exception:", err);
+  console.error(err.stack ?? "(no stack trace)");
+});
+
+process.on("unhandledRejection", (reason) => {
+  console.error("[FATAL] Unhandled rejection:", reason);
+  if (reason instanceof Error) {
+    console.error(reason.stack ?? "(no stack trace)");
+  }
+});
+
+import { writeFileSync, unlinkSync, mkdirSync, existsSync } from "node:fs";
+import { spawn, type ChildProcess } from "node:child_process";
 import { join } from "node:path";
 import { homedir } from "node:os";
 import { config } from "dotenv";
@@ -18,8 +33,64 @@ import { startDashboard } from "../dashboard/server.js";
 import { readEnv } from "./services/env.js";
 import { startTunnel, stopTunnel, isTunnelRunning, getTunnelUrl } from "./services/tunnel.js";
 import { startTwilioServer } from "./services/twilio-manager.js";
-import { startHeartbeat } from "./services/heartbeat.js";
-import { startVoiceServer } from "./voice/voice-server.js";
+
+/** Base URL for the Python FastAPI server (for tunnel URL notification) */
+const VOICE_SERVER_API_URL = process.env.VOICE_SERVER_URL ?? "http://localhost:7861";
+
+/** Path to the Python voice server directory */
+const VOICE_SERVER_DIR = join(import.meta.dirname ?? ".", "..", "voice-server");
+
+/** Reference to the Python voice server child process */
+let pythonProcess: ChildProcess | null = null;
+
+/**
+ * Start the Python voice server as a child process.
+ * Waits for the health endpoint to respond before returning.
+ */
+async function startPythonVoiceServer(): Promise<void> {
+  const venvPython = join(VOICE_SERVER_DIR, ".venv", "bin", "python");
+  if (!existsSync(venvPython)) {
+    console.warn(`Python venv not found at ${venvPython} -- voice server will not start`);
+    return;
+  }
+
+  console.log("Starting Python voice server...");
+  pythonProcess = spawn(venvPython, ["server.py"], {
+    cwd: VOICE_SERVER_DIR,
+    stdio: ["ignore", "inherit", "inherit"],
+  });
+
+  pythonProcess.on("exit", (code) => {
+    console.error(`Python voice server exited with code ${code}`);
+    pythonProcess = null;
+  });
+
+  // Wait for health endpoint (up to 15s)
+  const deadline = Date.now() + 15_000;
+  while (Date.now() < deadline) {
+    try {
+      const res = await fetch(`${VOICE_SERVER_API_URL}/health`);
+      if (res.ok) {
+        console.log("Python voice server is ready");
+        return;
+      }
+    } catch {
+      // Not ready yet
+    }
+    await new Promise((r) => setTimeout(r, 500));
+  }
+  console.warn("Python voice server did not become healthy within 15s -- continuing anyway");
+}
+
+/**
+ * Stop the Python voice server child process.
+ */
+function stopPythonVoiceServer(): void {
+  if (pythonProcess) {
+    pythonProcess.kill("SIGTERM");
+    pythonProcess = null;
+  }
+}
 
 // Use VOICECC_DIR env var if set (passed by CLI when dropping root privileges),
 // otherwise fall back to ~/.voicecc.
@@ -64,20 +135,36 @@ function cleanupStatusFile(): void {
 
 async function main(): Promise<void> {
   const dashboardPort = await startDashboard();
-  const voicePort = await startVoiceServer(dashboardPort);
 
-  startHeartbeat();
+  // Start the Python voice server (voice pipeline + text chat + heartbeat)
+  await startPythonVoiceServer();
 
   const envVars = await readEnv();
 
   // Write status file early so the CLI can show dashboard info while tunnel starts
   writeStatusFile(dashboardPort, null);
 
-  // Auto-start tunnel if enabled (independent of integrations)
+  // Auto-start tunnel if enabled -- tunnel now points at dashboard port
+  // so all external traffic goes through dashboard auth
   if (envVars.TUNNEL_ENABLED === "true") {
     try {
-      await startTunnel(voicePort);
-      writeStatusFile(dashboardPort, getTunnelUrl());
+      await startTunnel(dashboardPort);
+      const tunnelUrl = getTunnelUrl();
+      writeStatusFile(dashboardPort, tunnelUrl);
+
+      // Notify Python server of the tunnel URL so it can build TwiML URLs
+      if (tunnelUrl) {
+        try {
+          await fetch(`${VOICE_SERVER_API_URL}/config/tunnel-url`, {
+            method: "POST",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify({ url: tunnelUrl }),
+          });
+          console.log(`Notified Python server of tunnel URL: ${tunnelUrl}`);
+        } catch (notifyErr) {
+          console.warn(`Failed to notify Python server of tunnel URL: ${notifyErr}`);
+        }
+      }
     } catch (err) {
       const errorMsg = err instanceof Error ? err.message : String(err);
       console.error(`Tunnel auto-start failed: ${errorMsg}`);
@@ -101,6 +188,7 @@ async function main(): Promise<void> {
 
   // Graceful shutdown: stop tunnel subprocess, then clean up status file
   const shutdown = () => {
+    stopPythonVoiceServer();
     stopTunnel();
     cleanupStatusFile();
     process.exit(0);

package/server/services/device-pairing.ts

@@ -47,6 +47,7 @@ export interface PairingValidation {
 
 const PAIRING_CODE_TTL_MS = 5 * 60 * 1000;
 const PAIRING_MAX_ATTEMPTS = 5;
+const DEVICE_TOKEN_TTL_MS = 24 * 60 * 60 * 1000;
 const DEVICE_TOKENS_PATH = join(process.cwd(), ".device-tokens.json");
 
 // ============================================================================
@@ -141,7 +142,14 @@ export function isPairingCodeActive(code: string): boolean {
 }
 
 export function isValidDeviceToken(token: string): boolean {
-  return deviceTokens.has(token);
+  const info = deviceTokens.get(token);
+  if (!info) return false;
+  if (Date.now() - info.pairedAt > DEVICE_TOKEN_TTL_MS) {
+    deviceTokens.delete(token);
+    saveDeviceTokens().catch(() => {});
+    return false;
+  }
+  return true;
 }
 
 /**
@@ -151,9 +159,17 @@ export function isValidDeviceToken(token: string): boolean {
 export async function loadDeviceTokens(): Promise<void> {
   try {
     const data = JSON.parse(await readFile(DEVICE_TOKENS_PATH, "utf-8"));
+    const now = Date.now();
+    let pruned = false;
     for (const [token, info] of Object.entries(data)) {
-      deviceTokens.set(token, info as DeviceTokenInfo);
+      const typed = info as DeviceTokenInfo;
+      if (now - typed.pairedAt > DEVICE_TOKEN_TTL_MS) {
+        pruned = true;
+      } else {
+        deviceTokens.set(token, typed);
+      }
     }
+    if (pruned) saveDeviceTokens().catch(() => {});
   } catch {
     // File doesn't exist or is invalid -- start fresh
   }

package/server/services/twilio-manager.ts

@@ -1,14 +1,26 @@
 /**
  * Twilio integration state management.
  *
- * Tracks whether the Twilio integration is enabled/active and handles
- * Twilio-specific setup (webhook URL updates). The actual HTTP/WebSocket
- * handling runs in the unified voice server (voice-server.ts).
+ * Simplified: tracks whether Twilio is enabled and checks the Python voice
+ * server health. The actual Twilio call handling (WebSocket, TwiML, heartbeat)
+ * runs in the Python server.
+ *
+ * Responsibilities:
+ * - Track Twilio running state
+ * - Update Twilio phone number webhooks on start
+ * - Check Python server health via GET /health
  */
 
 import { readEnv } from "./env.js";
 import twilioSdk from "twilio";
 
+// ============================================================================
+// CONSTANTS
+// ============================================================================
+
+/** Base URL for the Python FastAPI server */
+const VOICE_API_URL = process.env.VOICE_SERVER_URL ?? "http://localhost:7861";
+
 // ============================================================================
 // TYPES
 // ============================================================================
@@ -22,7 +34,7 @@ export interface TwilioStatus {
 // STATE
 // ============================================================================
 
-/** Whether the Twilio voice server is running */
+/** Whether the Twilio integration is running */
 let twilioRunning = false;
 
 // ============================================================================
@@ -31,9 +43,7 @@ let twilioRunning = false;
 
 /**
  * Start the Twilio integration.
- * Reads .env for TWILIO_AUTH_TOKEN. If tunnelUrl exists, updates phone number
- * webhooks via Twilio SDK. The voice server is already running and handles
- * Twilio HTTP/WebSocket requests.
+ * Checks Python server health, then updates phone number webhooks via Twilio SDK.
  *
  * @param _dashboardPort - Unused (kept for API compatibility)
  * @param tunnelUrl - Optional tunnel public URL for webhook configuration
@@ -43,6 +53,17 @@ export async function startTwilioServer(_dashboardPort: number, tunnelUrl?: stri
     throw new Error("Twilio is already running");
   }
 
+  // Check Python server health
+  try {
+    const healthRes = await fetch(`${VOICE_API_URL}/health`);
+    if (!healthRes.ok) {
+      throw new Error(`Python server returned ${healthRes.status}`);
+    }
+  } catch (err) {
+    const msg = err instanceof Error ? err.message : String(err);
+    throw new Error(`Python voice server is not reachable at ${VOICE_API_URL}: ${msg}`);
+  }
+
   const envVars = await readEnv();
 
   if (!envVars.TWILIO_AUTH_TOKEN) {
@@ -50,7 +71,7 @@ export async function startTwilioServer(_dashboardPort: number, tunnelUrl?: stri
   }
 
   const accountSid = envVars.TWILIO_ACCOUNT_SID;
-  const webhookUrl = tunnelUrl ? `${tunnelUrl}/twilio/incoming-call` : null;
+  const webhookUrl = tunnelUrl ? `${tunnelUrl}/api/twilio/incoming-call` : null;
 
   if (tunnelUrl && accountSid && envVars.TWILIO_AUTH_TOKEN) {
     const client = twilioSdk(accountSid, envVars.TWILIO_AUTH_TOKEN);
@@ -80,8 +101,6 @@ export async function startTwilioServer(_dashboardPort: number, tunnelUrl?: stri
  * Stop the Twilio voice server.
  */
 export function stopTwilioServer(): void {
-  // In-process server doesn't have a clean shutdown mechanism yet;
-  // mark as not running so new calls are rejected.
   twilioRunning = false;
 }