@agenticmail/core 0.9.19 → 0.9.20

package/dist/index.cjs

@@ -7854,7 +7854,9 @@ var ElksRealtimeTransport = class {
   // Historical prefix — `elks-bye` / `elks-closed` etc. are matched by
   // long-standing call sites and tests; do not change.
   endReasonPrefix = "elks";
-  openaiAudioFormat = { type: "audio/pcm", rate: 24e3 };
+  // OpenAI rejects `format.rate` as an unknown parameter — `audio/pcm` is
+  // implicitly 24 kHz mono PCM16 in the current Realtime API.
+  openaiAudioFormat = { type: "audio/pcm" };
   parseInbound(raw) {
     const msg = parseElksRealtimeMessage(raw);
     if (msg.t === "hello") {
@@ -7885,9 +7887,9 @@ var TwilioRealtimeTransport = class {
   provider = "twilio";
   endReasonPrefix = "twilio";
   // µ-law @ 8 kHz — Twilio's native format; no transcode end to end.
-  // > `audio/pcmu` is the OpenAI GA Realtime µ-law format token; verify
-  // > against current OpenAI docs before the live smoke-test.
-  openaiAudioFormat = { type: "audio/pcmu", rate: 8e3 };
+  // OpenAI rejects `format.rate` as an unknown parameter; `audio/pcmu` is
+  // implicitly 8 kHz G.711 µ-law in the current Realtime API.
+  openaiAudioFormat = { type: "audio/pcmu" };
   /** Latched from the Twilio `start` frame; required on every outbound. */
   streamSid = "";
   /** The active `streamSid`, once the `start` frame has been seen. */
@@ -8247,7 +8249,7 @@ ${task}`);
   }
   return sections.join("\n\n");
 }
-var DEFAULT_REALTIME_AUDIO_FORMAT = { type: "audio/pcm", rate: REALTIME_AUDIO_SAMPLE_RATE };
+var DEFAULT_REALTIME_AUDIO_FORMAT = { type: "audio/pcm" };
 function buildRealtimeSessionConfig(opts) {
   const tools = opts.tools ?? [];
   const instructions = opts.instructions?.trim() || buildRealtimeInstructions({
@@ -8350,6 +8352,7 @@ var RealtimeVoiceBridge = class {
     if (this.ended || this.openaiReady) return;
     this.openaiReady = true;
     this.safeSend(this.openai, this.sessionConfig);
+    this.safeSend(this.openai, { type: "response.create" });
     for (const audio of this.pendingAudio.splice(0)) {
       this.safeSend(this.openai, { type: "input_audio_buffer.append", audio });
     }
@@ -9082,7 +9085,8 @@ function buildWebhookUrl(config, path2, missionId) {
   return url.toString();
 }
 function buildRealtimeStreamUrl(webhookBaseUrl, missionId, token) {
-  const url = new URL(`${apiBaseUrl(webhookBaseUrl)}${TWILIO_REALTIME_WS_PATH}`);
+  const root = webhookBaseUrl.replace(/\/+$/, "");
+  const url = new URL(`${root}${TWILIO_REALTIME_WS_PATH}`);
   url.protocol = url.protocol === "http:" ? "ws:" : "wss:";
   url.searchParams.set("missionId", missionId);
   url.searchParams.set("token", token);

package/dist/index.d.cts

@@ -2498,7 +2498,6 @@ declare class ElksRealtimeTransport implements RealtimeTransportAdapter {
     readonly endReasonPrefix = "elks";
     readonly openaiAudioFormat: {
         type: string;
-        rate: number;
     };
     constructor(listenFormat?: ElksRealtimeAudioFormat, sendFormat?: ElksRealtimeAudioFormat);
     parseInbound(raw: string | Record<string, unknown>): RealtimeInboundEvent;
@@ -2523,7 +2522,6 @@ declare class TwilioRealtimeTransport implements RealtimeTransportAdapter {
     readonly endReasonPrefix = "twilio";
     readonly openaiAudioFormat: {
         type: string;
-        rate: number;
     };
     /** Latched from the Twilio `start` frame; required on every outbound. */
     private streamSid;
@@ -2997,7 +2995,6 @@ interface RealtimeSessionConfigOptions extends RealtimeInstructionOptions {
 /** Default OpenAI Realtime audio format — linear PCM @ 24 kHz (46elks). */
 declare const DEFAULT_REALTIME_AUDIO_FORMAT: {
     readonly type: "audio/pcm";
-    readonly rate: 24000;
 };
 /**
  * Build the `session.update` client event for the GA `gpt-realtime`

package/dist/index.d.ts

@@ -2498,7 +2498,6 @@ declare class ElksRealtimeTransport implements RealtimeTransportAdapter {
     readonly endReasonPrefix = "elks";
     readonly openaiAudioFormat: {
         type: string;
-        rate: number;
     };
     constructor(listenFormat?: ElksRealtimeAudioFormat, sendFormat?: ElksRealtimeAudioFormat);
     parseInbound(raw: string | Record<string, unknown>): RealtimeInboundEvent;
@@ -2523,7 +2522,6 @@ declare class TwilioRealtimeTransport implements RealtimeTransportAdapter {
     readonly endReasonPrefix = "twilio";
     readonly openaiAudioFormat: {
         type: string;
-        rate: number;
     };
     /** Latched from the Twilio `start` frame; required on every outbound. */
     private streamSid;
@@ -2997,7 +2995,6 @@ interface RealtimeSessionConfigOptions extends RealtimeInstructionOptions {
 /** Default OpenAI Realtime audio format — linear PCM @ 24 kHz (46elks). */
 declare const DEFAULT_REALTIME_AUDIO_FORMAT: {
     readonly type: "audio/pcm";
-    readonly rate: 24000;
 };
 /**
  * Build the `session.update` client event for the GA `gpt-realtime`

package/dist/index.js

@@ -6942,7 +6942,9 @@ var ElksRealtimeTransport = class {
   // Historical prefix — `elks-bye` / `elks-closed` etc. are matched by
   // long-standing call sites and tests; do not change.
   endReasonPrefix = "elks";
-  openaiAudioFormat = { type: "audio/pcm", rate: 24e3 };
+  // OpenAI rejects `format.rate` as an unknown parameter — `audio/pcm` is
+  // implicitly 24 kHz mono PCM16 in the current Realtime API.
+  openaiAudioFormat = { type: "audio/pcm" };
   parseInbound(raw) {
     const msg = parseElksRealtimeMessage(raw);
     if (msg.t === "hello") {
@@ -6973,9 +6975,9 @@ var TwilioRealtimeTransport = class {
   provider = "twilio";
   endReasonPrefix = "twilio";
   // µ-law @ 8 kHz — Twilio's native format; no transcode end to end.
-  // > `audio/pcmu` is the OpenAI GA Realtime µ-law format token; verify
-  // > against current OpenAI docs before the live smoke-test.
-  openaiAudioFormat = { type: "audio/pcmu", rate: 8e3 };
+  // OpenAI rejects `format.rate` as an unknown parameter; `audio/pcmu` is
+  // implicitly 8 kHz G.711 µ-law in the current Realtime API.
+  openaiAudioFormat = { type: "audio/pcmu" };
   /** Latched from the Twilio `start` frame; required on every outbound. */
   streamSid = "";
   /** The active `streamSid`, once the `start` frame has been seen. */
@@ -7335,7 +7337,7 @@ ${task}`);
   }
   return sections.join("\n\n");
 }
-var DEFAULT_REALTIME_AUDIO_FORMAT = { type: "audio/pcm", rate: REALTIME_AUDIO_SAMPLE_RATE };
+var DEFAULT_REALTIME_AUDIO_FORMAT = { type: "audio/pcm" };
 function buildRealtimeSessionConfig(opts) {
   const tools = opts.tools ?? [];
   const instructions = opts.instructions?.trim() || buildRealtimeInstructions({
@@ -7438,6 +7440,7 @@ var RealtimeVoiceBridge = class {
     if (this.ended || this.openaiReady) return;
     this.openaiReady = true;
     this.safeSend(this.openai, this.sessionConfig);
+    this.safeSend(this.openai, { type: "response.create" });
     for (const audio of this.pendingAudio.splice(0)) {
       this.safeSend(this.openai, { type: "input_audio_buffer.append", audio });
     }
@@ -8170,7 +8173,8 @@ function buildWebhookUrl(config, path2, missionId) {
   return url.toString();
 }
 function buildRealtimeStreamUrl(webhookBaseUrl, missionId, token) {
-  const url = new URL(`${apiBaseUrl(webhookBaseUrl)}${TWILIO_REALTIME_WS_PATH}`);
+  const root = webhookBaseUrl.replace(/\/+$/, "");
+  const url = new URL(`${root}${TWILIO_REALTIME_WS_PATH}`);
   url.protocol = url.protocol === "http:" ? "ws:" : "wss:";
   url.searchParams.set("missionId", missionId);
   url.searchParams.set("token", token);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agenticmail/core",
-  "version": "0.9.19",
+  "version": "0.9.20",
   "description": "Core SDK for AgenticMail — email, SMS, and phone call-control for AI agents",
   "type": "module",
   "main": "dist/index.js",