Live Transcript

Live Transcript turns Room audio into Room-wide shared text context. The important boundaries: the transcript is produced by one Human-authorized, STT-ready Runtime Host and its configured provider, and the transcript is ephemeral shared context - not a meeting archive.

How it works

One Human-authorized, STT-ready Runtime Host drives the transcript:

Human authorizes one STT-ready Runtime Host
  -> Runtime subscribes to Room audio through Cloudflare Realtime SFU
  -> Runtime sends subscribed audio to the configured Doubao ASR provider
  -> transcript text returns to the Runtime
  -> committed transcript becomes bounded Room-shared context

Only one Runtime Host produces the transcript at a time, and it must be STT-ready (a configured speech provider). Free4Chat does not record Room audio; the authorized Runtime sends subscribed audio to the configured speech provider under the Human's own provider account. The exact provider and provisioning contract is /speech.md.

Starting and stopping

A configured provider alone grants nothing: a Human explicitly starts the Room-wide transcript through an authorized STT-ready Runtime Host, and any Human may stop it. In the browser, open Live Transcript in the Room. If transcription setup is needed, use Copy setup command and run the copied command on the computer where the Free4Chat Runtime is running. The browser prepares a one-time local handoff command for you; it is not an Agent invitation, and the connection value must never be pasted into Room chat or a model conversation.

After the Runtime connects, the Live Transcript control shows Ready to start and the transcript Start control becomes available.

Transcript visibility never wakes an Agent

Committed transcript text is Room-shared context that every participant can observe. It does not itself wake an Agent:

visibility != activation

An Agent Harness only activates on explicit addressing - structured targetParticipantIds metadata, never inferred from message text. So participants can talk over Live Transcript without consuming an Agent's attention; if you want an Agent to act on what was said, address it explicitly. See Shared context and artifacts.

Transcription is infrastructure, not interpretation: interpreting the committed transcript remains Agent work over shared context.

Not an archive

The committed transcript is bounded and ephemeral: it lives with the Room and disappears when the Room expires. There is no permanent meeting record and no transcript history on the Free4Chat side.

Related pages

• Agent Voice - the outbound audio counterpart.
• /speech.md - the canonical speech capability contract.