Prototype avatar popout local voice

[HashwanthVen]

Summary

Avatar-first popped-out agent window for the developer prototype.

• Routes popped-out agent windows to a dedicated avatar experience instead of the full chat UI (in Electron desktop only — browser mode falls back to the chat view).
• Adds a TalkingHead-based avatar stage with captions, microphone, and transcript exit controls.
• Keeps the main/default chat interface unchanged.
• Adds local developer voice support using Windows built-in speech recognition for STT and Edge TTS for speech output.

Updated for monorepo restructure

This PR was originally opened before master's monorepo restructure into apps/desktop, apps/web, apps/server, packages/services, packages/shared. The full prototype has been rebased onto the new layout:

Original (flat src/)	New location
src/main/ipc/voice.ts	apps/desktop/src/main/ipc/voice.ts
src/main.ts (wiring)	apps/desktop/src/main.ts (+ setupVoiceIPC())
src/preload.ts (voice namespace)	apps/desktop/src/preload.ts
src/main/ipc/mind.ts (popout window tweaks)	apps/desktop/src/main/ipc/mind.ts (860×900 + backgroundThrottling: false)
src/shared/types.ts (voice types)	packages/shared/src/types.ts
src/renderer/components/avatar/*	apps/web/src/renderer/components/avatar/*
src/renderer/types/talkinghead.d.ts	apps/web/src/renderer/types/talkinghead.d.ts
src/renderer/index.css (avatar stage CSS)	apps/web/src/renderer/index.css
src/renderer/components/layout/AppShell.tsx	apps/web/src/renderer/components/layout/AppShell.tsx (with browser-vs-Electron guard via window.electronAPI?.voice)
src/test/helpers.ts (mockElectronAPI voice)	apps/web/src/test/helpers.ts
vite.renderer.config.ts (optimizeDeps.exclude)	apps/web/vite.config.ts
src/main/services/sdk/sdkPaths.ts (dev-mode fix)	No longer needed — master's packages/services/src/sdk/sdkPaths.ts already enforces repo-local SDK in dev mode (more strictly than the original PR fix).

A new apps/web/src/browserApi.ts voice stub returns "unavailable in browser mode" errors so the browser variant doesn't crash if a developer opens the popout URL there. The avatar stage activates only when window.electronAPI.voice is present (Electron desktop).

Validation

• npm run lint — typecheck + eslint + dependency-cruiser clean (315 modules, 593 deps, 0 violations)
• npm test — full vitest suite clean (1050/1050 passing)
• Focused tests passing at new locations:
  • apps/web/src/renderer/components/avatar/PopoutAvatarWindow.test.tsx (4)
  • apps/web/src/renderer/components/avatar/AgentAvatarPanel.logic.test.ts (7)
  • packages/services/src/sdk/sdkPaths.test.ts (4)
  • packages/services/src/sdk/CopilotClientFactory.test.ts (8)
• Manual local dev smoke (npm start): avatar visible, mic captures speech via Windows System.Speech, agent reply plays via Edge TTS, red ✕ returns to transcript, main chat unchanged.

Dependencies added

• @met4citizen/talkinghead@^1.7.0 (MIT) → apps/web — avatar rendering
• edge-tts-universal@^1.4.0 (AGPL-3.0) → apps/desktop — local TTS output
  • Note: AGPL-3.0 is copyleft; flagged here for visibility. Used only at runtime in the desktop main process; not redistributed as source.

Follow-ups (intentionally out of scope here)

• Voice/avatar provider abstraction. Current implementation has flat voice IPC and a single TalkingHead path. Formalizing this behind a VoiceProvider interface should happen alongside any second provider (e.g. Foundry).
• Foundry / Azure Speech avatar integration as a separate optional provider — being explored on a separate spike branch, not mixed into the local prototype.
• Browser variant avatar/voice support — the popout currently degrades to the chat view in apps/web. Adding a Web Speech API + browser TTS shim could enable a browser-only path.

[HashwanthVen]

Hey @ianphil - Rebased onto current master, out of draft, all checks green.

What changed in this revision

Ported the prototype onto the new monorepo layout (apps/desktop, apps/web, packages/services, packages/shared). Same TalkingHead avatar + Windows System.Speech STT + Edge TTS - just at the right paths and importing through @chamber/shared / @chamber/services.

A couple things worth flagging:

• Added a small browser-mode voice stub in apps/web/browserApi.ts so the popout degrades gracefully (chat view) when window.electronAPI.voice isn't present.
• Dropped the original sdkPaths.ts tweak - your packages/services/src/sdk/sdkPaths.ts already enforces repo-local SDK in dev, more strictly than my fix did.
• 2 new deps: @met4citizen/talkinghead@^1.7.0 (MIT, apps/web) and edge-tts-universal@^1.4.0 (AGPL-3.0, apps/desktop - flagged in PR body, runtime-only in main process).

Validation

npm run lint clean, npm test 1050/1050 passing, CI green, local smoke tested (avatar + mic + TTS + interrupt + main chat all working).

Follow-ups (intentionally out of scope)

Foundry/Speech avatar will land as a separate optional provider per our chat. Voice provider abstraction deferred until the second provider exists. Also naturally absorbs #19 (Whisper Light onboarding voice) which you closed as superseded - happy to split that out if you'd prefer something narrower.

Let me know if you'd like me to break anything out, tighten scope, or address review comments - happy to iterate. 🎙️

[HashwanthVen]

One more thing worth surfacing while we're on this topic, Ian - I have an Azure Foundry-based avatar/speech version sitting on a separate fork branch that's the natural next step beyond this local prototype. Posting it here in case you want to take a look so we can decide how (or whether) to bring it into master.

Branch: https://github.com/HashwanthVen/chamber/tree/voice-avatar-foundry