feat: add audio transcription pipeline for voice messages

Adds Whisper-compatible audio transcription via configurable endpoint.
New functions: isSupportedAudio(), mimeToExtension(), transcribeAudio(),
buildUserMessageWithAudio(). Config schema gains audio section with
transcription_endpoint, api_key, and model. Daemon wires transcription
into the message router. Channel adapters extract audio from voice/audio
messages (Telegram voice+audio, Discord audio/*, Slack audio/*, WhatsApp
ptt+audio). Includes 57 media tests (was 25, now covers all audio paths).

src/daemon/index.ts

@@ -1,11 +1,15 @@
 import { Lifecycle } from './lifecycle.js';
 import type { Config, ModelConfig } from '../config/index.js';
+import type { AudioTranscriptionConfig } from '../models/media.js';
+import type { Attachment } from '../channels/types.js';
+import { isSupportedAudio, transcribeAudio } from '../models/media.js';
 import { AnthropicClient, OpenAIClient, OllamaClient, LlamaCppClient, GeminiClient, BedrockClient, GitHubModelsClient, ModelRouter, DEFAULT_RETRY_CONFIG } from '../models/index.js';
 import type { ModelClient, RetryConfig } from '../models/index.js';
 import { AgentOrchestrator, type DelegationConfig } from '../backends/index.js';
+import { OutboundAttachmentCollector } from '../backends/native/attachments.js';
 import { SessionStore, SessionManager } from '../session/index.js';
 import { HookEngine } from '../hooks/index.js';
-import { ToolRegistry, ToolExecutor, ToolPolicy, allBuiltinTools, createWebSearchTools, createProcessTools, ProcessManager, BrowserManager, createBrowserTools } from '../tools/index.js';
+import { ToolRegistry, ToolExecutor, ToolPolicy, allBuiltinTools, createWebSearchTools, createProcessTools, ProcessManager, BrowserManager, createBrowserTools, createMediaSendTool } from '../tools/index.js';
 import type { Tool } from '../tools/types.js';
 import { MemoryStore } from '../memory/index.js';
 import { createMemoryTools } from '../tools/builtin/index.js';
@@ -204,11 +208,12 @@ function createMessageRouter(deps: {
   agentConfigRegistry?: AgentConfigRegistry;
   agentRouter?: AgentRouter;
   sandboxManager?: SandboxManager;
+  audioConfig?: AudioTranscriptionConfig;
 }) {
   // Cache agents by session ID + agent config name to avoid recreating on every message
-  const agents = new Map<string, AgentOrchestrator>();
+  const agents = new Map<string, { orchestrator: AgentOrchestrator; collector: OutboundAttachmentCollector }>();
 
-  function getOrCreateAgent(channel: string, senderId: string): AgentOrchestrator {
+  function getOrCreateAgent(channel: string, senderId: string): { orchestrator: AgentOrchestrator; collector: OutboundAttachmentCollector } {
     // Resolve agent config name via routing (sender → channel → default fallback)
     const agentConfigName = deps.agentRouter?.resolve(channel, senderId);
     const agentConfig = agentConfigName ? deps.agentConfigRegistry?.get(agentConfigName) : undefined;
@@ -218,8 +223,8 @@ function createMessageRouter(deps: {
       ? `${channel}:${senderId}:${agentConfigName}`
       : `${channel}:${senderId}`;
 
-    let agent = agents.get(sessionId);
-    if (!agent) {
+    let entry = agents.get(sessionId);
+    if (!entry) {
       const session = deps.sessionManager.getSession(channel, senderId);
 
       // Use agent config overrides where available, falling back to global config
@@ -286,7 +291,14 @@ function createMessageRouter(deps: {
         effectiveToolRegistry.replace(lazySandboxProcess);
       }
 
-      agent = new AgentOrchestrator({
+      // Create an attachment collector for this agent session
+      const collector = new OutboundAttachmentCollector();
+
+      // Clone the tool registry to register the media.send tool bound to this collector
+      effectiveToolRegistry = effectiveToolRegistry.clone();
+      effectiveToolRegistry.register(createMediaSendTool(collector));
+
+      const orchestrator = new AgentOrchestrator({
         modelRouter: deps.modelRouter,
         systemPrompt: effectiveSystemPrompt,
         session,
@@ -307,14 +319,16 @@ function createMessageRouter(deps: {
           agent: effectiveTier,
           provider: effectiveProvider,
         },
+        attachmentCollector: collector,
       });
-      agents.set(sessionId, agent);
+      entry = { orchestrator, collector };
+      agents.set(sessionId, entry);
     }
-    return agent;
+    return entry;
   }
 
   return async (msg: InboundMessage, reply: (response: OutboundMessage) => Promise<void>): Promise<void> => {
-    const agent = getOrCreateAgent(msg.channel, msg.senderId);
+    const { orchestrator: agent, collector } = getOrCreateAgent(msg.channel, msg.senderId);
 
     // Handle special commands
     if (msg.metadata?.isCommand) {
@@ -367,8 +381,24 @@ function createMessageRouter(deps: {
     }
 
     try {
-      const response = await agent.process(msg.text, msg.attachments);
-      await reply({ text: response, replyTo: msg.id });
+      // Transcribe audio attachments before processing
+      let messageText = msg.text;
+      const audioAttachments = (msg.attachments ?? []).filter((a: Attachment) => isSupportedAudio(a));
+
+      if (audioAttachments.length > 0 && deps.audioConfig) {
+        for (const att of audioAttachments) {
+          const transcript = await transcribeAudio(att, deps.audioConfig);
+          messageText = `[Voice message]: ${transcript}\n\n${messageText}`;
+        }
+      }
+
+      const response = await agent.process(messageText, msg.attachments);
+      const outboundAttachments = collector.drain();
+      await reply({
+        text: response,
+        replyTo: msg.id,
+        attachments: outboundAttachments.length > 0 ? outboundAttachments : undefined,
+      });
     } catch (error) {
       console.error(`Error processing message from ${msg.channel}:${msg.senderId}:`, error);
       await reply({
@@ -539,6 +569,13 @@ export async function startDaemon(config: Config): Promise<DaemonContext> {
     });
   }
 
+  // Initialize audio transcription config
+  const audioConfig: AudioTranscriptionConfig = {
+    endpoint: config.audio.transcription_endpoint,
+    apiKey: config.audio.transcription_api_key,
+    model: config.audio.transcription_model,
+  };
+
   // Initialize model router
   const modelRouter = createModelRouter(config);
 
@@ -593,6 +630,7 @@ export async function startDaemon(config: Config): Promise<DaemonContext> {
     agentConfigRegistry,
     agentRouter,
     sandboxManager,
+    audioConfig,
   }));
 
   // Register Telegram adapter