488 lines
16 KiB
TypeScript
488 lines
16 KiB
TypeScript
import http from "node:http";
|
|
import { URL } from "node:url";
|
|
import {
|
|
isRequestBodyLimitError,
|
|
readRequestBodyWithLimit,
|
|
requestBodyErrorToText,
|
|
} from "openclaw/plugin-sdk/voice-call";
|
|
import { normalizeVoiceCallConfig, type VoiceCallConfig } from "./config.js";
|
|
import type { CoreConfig } from "./core-bridge.js";
|
|
import type { CallManager } from "./manager.js";
|
|
import type { MediaStreamConfig } from "./media-stream.js";
|
|
import { MediaStreamHandler } from "./media-stream.js";
|
|
import type { VoiceCallProvider } from "./providers/base.js";
|
|
import { OpenAIRealtimeSTTProvider } from "./providers/stt-openai-realtime.js";
|
|
import type { TwilioProvider } from "./providers/twilio.js";
|
|
import type { NormalizedEvent, WebhookContext } from "./types.js";
|
|
import { startStaleCallReaper } from "./webhook/stale-call-reaper.js";
|
|
|
|
const MAX_WEBHOOK_BODY_BYTES = 1024 * 1024;
|
|
|
|
type WebhookResponsePayload = {
|
|
statusCode: number;
|
|
body: string;
|
|
headers?: Record<string, string>;
|
|
};
|
|
|
|
function buildRequestUrl(
|
|
requestUrl: string | undefined,
|
|
requestHost: string | undefined,
|
|
fallbackHost = "localhost",
|
|
): URL {
|
|
return new URL(requestUrl ?? "/", `http://${requestHost ?? fallbackHost}`);
|
|
}
|
|
|
|
function normalizeWebhookResponse(parsed: {
|
|
statusCode?: number;
|
|
providerResponseHeaders?: Record<string, string>;
|
|
providerResponseBody?: string;
|
|
}): WebhookResponsePayload {
|
|
return {
|
|
statusCode: parsed.statusCode ?? 200,
|
|
headers: parsed.providerResponseHeaders,
|
|
body: parsed.providerResponseBody ?? "OK",
|
|
};
|
|
}
|
|
|
|
/**
|
|
* HTTP server for receiving voice call webhooks from providers.
|
|
* Supports WebSocket upgrades for media streams when streaming is enabled.
|
|
*/
|
|
export class VoiceCallWebhookServer {
|
|
private server: http.Server | null = null;
|
|
private listeningUrl: string | null = null;
|
|
private config: VoiceCallConfig;
|
|
private manager: CallManager;
|
|
private provider: VoiceCallProvider;
|
|
private coreConfig: CoreConfig | null;
|
|
private stopStaleCallReaper: (() => void) | null = null;
|
|
|
|
/** Media stream handler for bidirectional audio (when streaming enabled) */
|
|
private mediaStreamHandler: MediaStreamHandler | null = null;
|
|
|
|
constructor(
|
|
config: VoiceCallConfig,
|
|
manager: CallManager,
|
|
provider: VoiceCallProvider,
|
|
coreConfig?: CoreConfig,
|
|
) {
|
|
this.config = normalizeVoiceCallConfig(config);
|
|
this.manager = manager;
|
|
this.provider = provider;
|
|
this.coreConfig = coreConfig ?? null;
|
|
|
|
// Initialize media stream handler if streaming is enabled
|
|
if (this.config.streaming.enabled) {
|
|
this.initializeMediaStreaming();
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Get the media stream handler (for wiring to provider).
|
|
*/
|
|
getMediaStreamHandler(): MediaStreamHandler | null {
|
|
return this.mediaStreamHandler;
|
|
}
|
|
|
|
/**
|
|
* Initialize media streaming with OpenAI Realtime STT.
|
|
*/
|
|
private initializeMediaStreaming(): void {
|
|
const streaming = this.config.streaming;
|
|
const apiKey = streaming.openaiApiKey ?? process.env.OPENAI_API_KEY;
|
|
|
|
if (!apiKey) {
|
|
console.warn("[voice-call] Streaming enabled but no OpenAI API key found");
|
|
return;
|
|
}
|
|
|
|
const sttProvider = new OpenAIRealtimeSTTProvider({
|
|
apiKey,
|
|
model: streaming.sttModel,
|
|
silenceDurationMs: streaming.silenceDurationMs,
|
|
vadThreshold: streaming.vadThreshold,
|
|
});
|
|
|
|
const streamConfig: MediaStreamConfig = {
|
|
sttProvider,
|
|
preStartTimeoutMs: streaming.preStartTimeoutMs,
|
|
maxPendingConnections: streaming.maxPendingConnections,
|
|
maxPendingConnectionsPerIp: streaming.maxPendingConnectionsPerIp,
|
|
maxConnections: streaming.maxConnections,
|
|
shouldAcceptStream: ({ callId, token }) => {
|
|
const call = this.manager.getCallByProviderCallId(callId);
|
|
if (!call) {
|
|
return false;
|
|
}
|
|
if (this.provider.name === "twilio") {
|
|
const twilio = this.provider as TwilioProvider;
|
|
if (!twilio.isValidStreamToken(callId, token)) {
|
|
console.warn(`[voice-call] Rejecting media stream: invalid token for ${callId}`);
|
|
return false;
|
|
}
|
|
}
|
|
return true;
|
|
},
|
|
onTranscript: (providerCallId, transcript) => {
|
|
console.log(`[voice-call] Transcript for ${providerCallId}: ${transcript}`);
|
|
|
|
// Clear TTS queue on barge-in (user started speaking, interrupt current playback)
|
|
if (this.provider.name === "twilio") {
|
|
(this.provider as TwilioProvider).clearTtsQueue(providerCallId);
|
|
}
|
|
|
|
// Look up our internal call ID from the provider call ID
|
|
const call = this.manager.getCallByProviderCallId(providerCallId);
|
|
if (!call) {
|
|
console.warn(`[voice-call] No active call found for provider ID: ${providerCallId}`);
|
|
return;
|
|
}
|
|
|
|
// Create a speech event and process it through the manager
|
|
const event: NormalizedEvent = {
|
|
id: `stream-transcript-${Date.now()}`,
|
|
type: "call.speech",
|
|
callId: call.callId,
|
|
providerCallId,
|
|
timestamp: Date.now(),
|
|
transcript,
|
|
isFinal: true,
|
|
};
|
|
this.manager.processEvent(event);
|
|
|
|
// Auto-respond in conversation mode (inbound always, outbound if mode is conversation)
|
|
const callMode = call.metadata?.mode as string | undefined;
|
|
const shouldRespond = call.direction === "inbound" || callMode === "conversation";
|
|
if (shouldRespond) {
|
|
this.handleInboundResponse(call.callId, transcript).catch((err) => {
|
|
console.warn(`[voice-call] Failed to auto-respond:`, err);
|
|
});
|
|
}
|
|
},
|
|
onSpeechStart: (providerCallId) => {
|
|
if (this.provider.name === "twilio") {
|
|
(this.provider as TwilioProvider).clearTtsQueue(providerCallId);
|
|
}
|
|
},
|
|
onPartialTranscript: (callId, partial) => {
|
|
console.log(`[voice-call] Partial for ${callId}: ${partial}`);
|
|
},
|
|
onConnect: (callId, streamSid) => {
|
|
console.log(`[voice-call] Media stream connected: ${callId} -> ${streamSid}`);
|
|
// Register stream with provider for TTS routing
|
|
if (this.provider.name === "twilio") {
|
|
(this.provider as TwilioProvider).registerCallStream(callId, streamSid);
|
|
}
|
|
|
|
// Speak initial message if one was provided when call was initiated
|
|
// Use setTimeout to allow stream setup to complete
|
|
setTimeout(() => {
|
|
this.manager.speakInitialMessage(callId).catch((err) => {
|
|
console.warn(`[voice-call] Failed to speak initial message:`, err);
|
|
});
|
|
}, 500);
|
|
},
|
|
onDisconnect: (callId) => {
|
|
console.log(`[voice-call] Media stream disconnected: ${callId}`);
|
|
// Auto-end call when media stream disconnects to prevent stuck calls.
|
|
// Without this, calls can remain active indefinitely after the stream closes.
|
|
const disconnectedCall = this.manager.getCallByProviderCallId(callId);
|
|
if (disconnectedCall) {
|
|
console.log(
|
|
`[voice-call] Auto-ending call ${disconnectedCall.callId} on stream disconnect`,
|
|
);
|
|
void this.manager.endCall(disconnectedCall.callId).catch((err) => {
|
|
console.warn(`[voice-call] Failed to auto-end call ${disconnectedCall.callId}:`, err);
|
|
});
|
|
}
|
|
if (this.provider.name === "twilio") {
|
|
(this.provider as TwilioProvider).unregisterCallStream(callId);
|
|
}
|
|
},
|
|
};
|
|
|
|
this.mediaStreamHandler = new MediaStreamHandler(streamConfig);
|
|
console.log("[voice-call] Media streaming initialized");
|
|
}
|
|
|
|
/**
|
|
* Start the webhook server.
|
|
* Idempotent: returns immediately if the server is already listening.
|
|
*/
|
|
async start(): Promise<string> {
|
|
const { port, bind, path: webhookPath } = this.config.serve;
|
|
const streamPath = this.config.streaming.streamPath;
|
|
|
|
// Guard: if a server is already listening, return the existing URL.
|
|
// This prevents EADDRINUSE when start() is called more than once on the
|
|
// same instance (e.g. during config hot-reload or concurrent ensureRuntime).
|
|
if (this.server?.listening) {
|
|
return this.listeningUrl ?? this.resolveListeningUrl(bind, webhookPath);
|
|
}
|
|
|
|
return new Promise((resolve, reject) => {
|
|
this.server = http.createServer((req, res) => {
|
|
this.handleRequest(req, res, webhookPath).catch((err) => {
|
|
console.error("[voice-call] Webhook error:", err);
|
|
res.statusCode = 500;
|
|
res.end("Internal Server Error");
|
|
});
|
|
});
|
|
|
|
// Handle WebSocket upgrades for media streams
|
|
if (this.mediaStreamHandler) {
|
|
this.server.on("upgrade", (request, socket, head) => {
|
|
const path = this.getUpgradePathname(request);
|
|
if (path === streamPath) {
|
|
console.log("[voice-call] WebSocket upgrade for media stream");
|
|
this.mediaStreamHandler?.handleUpgrade(request, socket, head);
|
|
} else {
|
|
socket.destroy();
|
|
}
|
|
});
|
|
}
|
|
|
|
this.server.on("error", reject);
|
|
|
|
this.server.listen(port, bind, () => {
|
|
const url = this.resolveListeningUrl(bind, webhookPath);
|
|
this.listeningUrl = url;
|
|
console.log(`[voice-call] Webhook server listening on ${url}`);
|
|
if (this.mediaStreamHandler) {
|
|
const address = this.server?.address();
|
|
const actualPort =
|
|
address && typeof address === "object" ? address.port : this.config.serve.port;
|
|
console.log(
|
|
`[voice-call] Media stream WebSocket on ws://${bind}:${actualPort}${streamPath}`,
|
|
);
|
|
}
|
|
resolve(url);
|
|
|
|
// Start the stale call reaper if configured
|
|
this.stopStaleCallReaper = startStaleCallReaper({
|
|
manager: this.manager,
|
|
staleCallReaperSeconds: this.config.staleCallReaperSeconds,
|
|
});
|
|
});
|
|
});
|
|
}
|
|
|
|
/**
|
|
* Stop the webhook server.
|
|
*/
|
|
async stop(): Promise<void> {
|
|
if (this.stopStaleCallReaper) {
|
|
this.stopStaleCallReaper();
|
|
this.stopStaleCallReaper = null;
|
|
}
|
|
return new Promise((resolve) => {
|
|
if (this.server) {
|
|
this.server.close(() => {
|
|
this.server = null;
|
|
this.listeningUrl = null;
|
|
resolve();
|
|
});
|
|
} else {
|
|
this.listeningUrl = null;
|
|
resolve();
|
|
}
|
|
});
|
|
}
|
|
|
|
private resolveListeningUrl(bind: string, webhookPath: string): string {
|
|
const address = this.server?.address();
|
|
if (address && typeof address === "object") {
|
|
const host = address.address && address.address.length > 0 ? address.address : bind;
|
|
const normalizedHost = host.includes(":") && !host.startsWith("[") ? `[${host}]` : host;
|
|
return `http://${normalizedHost}:${address.port}${webhookPath}`;
|
|
}
|
|
return `http://${bind}:${this.config.serve.port}${webhookPath}`;
|
|
}
|
|
|
|
private getUpgradePathname(request: http.IncomingMessage): string | null {
|
|
try {
|
|
return buildRequestUrl(request.url, request.headers.host).pathname;
|
|
} catch {
|
|
return null;
|
|
}
|
|
}
|
|
|
|
private normalizeWebhookPathForMatch(pathname: string): string {
|
|
const trimmed = pathname.trim();
|
|
if (!trimmed) {
|
|
return "/";
|
|
}
|
|
const prefixed = trimmed.startsWith("/") ? trimmed : `/${trimmed}`;
|
|
if (prefixed === "/") {
|
|
return prefixed;
|
|
}
|
|
return prefixed.endsWith("/") ? prefixed.slice(0, -1) : prefixed;
|
|
}
|
|
|
|
private isWebhookPathMatch(requestPath: string, configuredPath: string): boolean {
|
|
return (
|
|
this.normalizeWebhookPathForMatch(requestPath) ===
|
|
this.normalizeWebhookPathForMatch(configuredPath)
|
|
);
|
|
}
|
|
|
|
/**
|
|
* Handle incoming HTTP request.
|
|
*/
|
|
private async handleRequest(
|
|
req: http.IncomingMessage,
|
|
res: http.ServerResponse,
|
|
webhookPath: string,
|
|
): Promise<void> {
|
|
const payload = await this.runWebhookPipeline(req, webhookPath);
|
|
this.writeWebhookResponse(res, payload);
|
|
}
|
|
|
|
private async runWebhookPipeline(
|
|
req: http.IncomingMessage,
|
|
webhookPath: string,
|
|
): Promise<WebhookResponsePayload> {
|
|
const url = buildRequestUrl(req.url, req.headers.host);
|
|
|
|
if (url.pathname === "/voice/hold-music") {
|
|
return {
|
|
statusCode: 200,
|
|
headers: { "Content-Type": "text/xml" },
|
|
body: `<?xml version="1.0" encoding="UTF-8"?>
|
|
<Response>
|
|
<Say voice="alice">All agents are currently busy. Please hold.</Say>
|
|
<Play loop="0">https://s3.amazonaws.com/com.twilio.music.classical/BusyStrings.mp3</Play>
|
|
</Response>`,
|
|
};
|
|
}
|
|
|
|
if (!this.isWebhookPathMatch(url.pathname, webhookPath)) {
|
|
return { statusCode: 404, body: "Not Found" };
|
|
}
|
|
|
|
if (req.method !== "POST") {
|
|
return { statusCode: 405, body: "Method Not Allowed" };
|
|
}
|
|
|
|
let body = "";
|
|
try {
|
|
body = await this.readBody(req, MAX_WEBHOOK_BODY_BYTES);
|
|
} catch (err) {
|
|
if (isRequestBodyLimitError(err, "PAYLOAD_TOO_LARGE")) {
|
|
return { statusCode: 413, body: "Payload Too Large" };
|
|
}
|
|
if (isRequestBodyLimitError(err, "REQUEST_BODY_TIMEOUT")) {
|
|
return { statusCode: 408, body: requestBodyErrorToText("REQUEST_BODY_TIMEOUT") };
|
|
}
|
|
throw err;
|
|
}
|
|
|
|
const ctx: WebhookContext = {
|
|
headers: req.headers as Record<string, string | string[] | undefined>,
|
|
rawBody: body,
|
|
url: url.toString(),
|
|
method: "POST",
|
|
query: Object.fromEntries(url.searchParams),
|
|
remoteAddress: req.socket.remoteAddress ?? undefined,
|
|
};
|
|
|
|
const verification = this.provider.verifyWebhook(ctx);
|
|
if (!verification.ok) {
|
|
console.warn(`[voice-call] Webhook verification failed: ${verification.reason}`);
|
|
return { statusCode: 401, body: "Unauthorized" };
|
|
}
|
|
if (!verification.verifiedRequestKey) {
|
|
console.warn("[voice-call] Webhook verification succeeded without request identity key");
|
|
return { statusCode: 401, body: "Unauthorized" };
|
|
}
|
|
|
|
const parsed = this.provider.parseWebhookEvent(ctx, {
|
|
verifiedRequestKey: verification.verifiedRequestKey,
|
|
});
|
|
|
|
if (verification.isReplay) {
|
|
console.warn("[voice-call] Replay detected; skipping event side effects");
|
|
} else {
|
|
this.processParsedEvents(parsed.events);
|
|
}
|
|
|
|
return normalizeWebhookResponse(parsed);
|
|
}
|
|
|
|
private processParsedEvents(events: NormalizedEvent[]): void {
|
|
for (const event of events) {
|
|
try {
|
|
this.manager.processEvent(event);
|
|
} catch (err) {
|
|
console.error(`[voice-call] Error processing event ${event.type}:`, err);
|
|
}
|
|
}
|
|
}
|
|
|
|
private writeWebhookResponse(res: http.ServerResponse, payload: WebhookResponsePayload): void {
|
|
res.statusCode = payload.statusCode;
|
|
if (payload.headers) {
|
|
for (const [key, value] of Object.entries(payload.headers)) {
|
|
res.setHeader(key, value);
|
|
}
|
|
}
|
|
res.end(payload.body);
|
|
}
|
|
|
|
/**
|
|
* Read request body as string with timeout protection.
|
|
*/
|
|
private readBody(
|
|
req: http.IncomingMessage,
|
|
maxBytes: number,
|
|
timeoutMs = 30_000,
|
|
): Promise<string> {
|
|
return readRequestBodyWithLimit(req, { maxBytes, timeoutMs });
|
|
}
|
|
|
|
/**
|
|
* Handle auto-response for inbound calls using the agent system.
|
|
* Supports tool calling for richer voice interactions.
|
|
*/
|
|
private async handleInboundResponse(callId: string, userMessage: string): Promise<void> {
|
|
console.log(`[voice-call] Auto-responding to inbound call ${callId}: "${userMessage}"`);
|
|
|
|
// Get call context for conversation history
|
|
const call = this.manager.getCall(callId);
|
|
if (!call) {
|
|
console.warn(`[voice-call] Call ${callId} not found for auto-response`);
|
|
return;
|
|
}
|
|
|
|
if (!this.coreConfig) {
|
|
console.warn("[voice-call] Core config missing; skipping auto-response");
|
|
return;
|
|
}
|
|
|
|
try {
|
|
const { generateVoiceResponse } = await import("./response-generator.js");
|
|
|
|
const result = await generateVoiceResponse({
|
|
voiceConfig: this.config,
|
|
coreConfig: this.coreConfig,
|
|
callId,
|
|
from: call.from,
|
|
transcript: call.transcript,
|
|
userMessage,
|
|
});
|
|
|
|
if (result.error) {
|
|
console.error(`[voice-call] Response generation error: ${result.error}`);
|
|
return;
|
|
}
|
|
|
|
if (result.text) {
|
|
console.log(`[voice-call] AI response: "${result.text}"`);
|
|
await this.manager.speak(callId, result.text);
|
|
}
|
|
} catch (err) {
|
|
console.error(`[voice-call] Auto-response error:`, err);
|
|
}
|
|
}
|
|
}
|