Voice mode

Voice mode replaces the web chat's messages and composer with a speaking orb. An adjacent page or site preview stays open. The selected agent, model, page context, tools, permissions, and message queue remain the same as in text chat.

Coding agents keep running Claude Code in their existing sandbox. Voice mode does not change the chat's pinned runtime, checkout, preview, or Claude session. The voice response style travels with each turn because Claude Code restores the original system prompt when resuming a session. Text turns explicitly restore normal response style. Decopilot chats use the same speech transport.

Choose Start voice mode beside the composer and allow microphone access. Back to chat stops the microphone and playback. It preserves the chat and any running work. Stop agent work is a separate control. Speaking while a reply plays interrupts the audio; the new request joins the chat's normal queue. Approvals, questions, errors, and queued requests remain visible in voice mode.

Only the final text part of the current spoken turn is read aloud. A per-turn instruction asks the agent for a short, natural answer. Switching back to text restores normal response style for subsequent messages.

Deployment setup

Set these variables on every API replica:

ELEVENLABS_API_KEY=<elevenlabs-api-key>

The key needs Text to Speech and Speech to Text access. Keep it on the server. Browsers receive a single-use Scribe token and a signed Studio session token. The web app needs HTTPS for microphone access, except on localhost. No inbound provider callback or Speech Engine provisioning is required. API replicas must share the same Studio authentication secret and NATS service.

Then enable Settings → General → Voice mode for the organization. The voice_mode organization flag defaults off. It is separate from authentication: only the owning user can start a voice session for a writable, hosted chat. Native terminal chats and read-only chats do not offer this mode.

Optional variables:

Session behavior and limits

The browser closes sessions after ten minutes, with one active session per user per organization. Server speech grants expire after ten minutes and allow up to 120 synthesis requests and 24,000 characters. An interrupted browser that cannot release its reservation may need to wait for that expiry before reconnecting. ElevenLabs bills speech usage to the deployment's account; Studio's model usage accounting continues to cover the agent's model calls.

Microphone audio travels directly from the browser to ElevenLabs Scribe Realtime. Voice activity detection commits the utterance after a pause. The browser submits it through the existing chat API and waits for the agent to finish. Studio then synthesizes the final answer with ElevenLabs v4 Turbo and plays it in the browser. Transcription and synthesis are independent so a long-running coding task does not hit the Speech Engine's response timeout.

Transcripts and answers remain in the normal Studio chat history. Studio does not store audio; ElevenLabs' data retention policy applies to its service.

Leaving the chat, changing organizations, or disconnecting stops the voice session. Transport failures show an error with a path back to text. They do not automatically resend a request whose acceptance is uncertain. Check the chat before repeating it. Stopping playback leaves the agent's work running.

Verification

The regular voice-mode.spec.ts browser suite checks access gates, the default flag, and switching back to a preserved draft. Its optional live case requires an API server configured with ElevenLabs and a synthetic WAV input:

E2E_VOICE_LIVE=1 E2E_VOICE_WAV=/tmp/synthetic-utterance.wav \
  bun run --cwd=packages/e2e test:e2e -- tests/voice-mode.spec.ts

The live case uses the real speech service and an HTTP model stand-in, and checks that the voice instruction reaches the agent and the reply is spoken.

The daemon suite also exercises the real Claude Code executable and SDK against an HTTP model stand-in. It starts in text, changes a file in voice mode, and returns to text while checking the same Claude session and conversation history:

DAEMON_E2E_CLAUDE_VOICE=1 \
  bun test packages/sandbox/daemon-e2e/daemon.voice-claude.e2e.test.ts

Build the Go daemon first, as described in the sandbox package README. Set DAEMON_E2E_CMD to use a daemon binary at another path.