Repository navigation
berd-call: stream realtime calls live or at handoffs - #390
Closed
johnmatthewtennant wants to merge 1 commit into
Closed
johnmatthewtennant wants to merge 1 commit into
johnmatthewtennant wants to merge 1 commit into
Conversation
An Expert driving a realtime call from the CLI should hear the conversation the way a GPT-Live delegate does: only when the Spokesperson hands off, with everything said since the last handoff. Until now `start --stream` and `--codex` also flushed after every Spokesperson reply, waking the Expert on each turn, while a host that shows the call as it happens saw the user's words only once the reply had played. Bare `--stream` (or `--stream handoffs`) and `--codex` now hold rows until a handoff. `--stream live` shows the conversation as it happens: a `user` row as soon as the words are transcribed, and a `spokesperson_generated` row as soon as a reply is generated, before it is spoken and only once no user speech is still being transcribed. The session emits these as `user_transcribed` and `spokesperson_generated` previews; the transcript recorded after playback stays authoritative. Co-Authored-By: Claude Opus 5.5 <[email protected]>
Contributor
Author
|
🤖 Combined into #397 with its commits kept, so the stack no longer needs restacking and review covers one diff. Open review findings carry over there. (AI-generated comment) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
An Expert driving a realtime (Expert-Spokesperson) call from the CLI should hear the conversation the way a GPT-Live client delegate does: only when the Spokesperson hands off, with everything said since the last handoff. Until now
start --streamand--codexalso flushed after every Spokesperson reply, waking the Expert on each turn. Meanwhile a host that shows the call as it happens (the Buzz Berd Call plugin) saw the user's words only after the reply had played, and the Spokesperson's reply only after playback finished.berd-call startnow has two stream modes for Expert-Spokesperson calls:--stream,--stream handoffs, and--codex): no rows until the Spokesperson hands off, then every row since the last handoff, ending with thehandoffrow.--stream live): auserrow as soon as the words are transcribed, and aspokesperson_generatedrow (cursor0) as soon as a reply is generated, before it is spoken. A generated reply waits while user speech is still being transcribed, so the stream reads in order. Thespokespersonorspokesperson_interruptedrow after playback still carries the cursor and what was heard, and user rows already shown are left out of the Expert batch.Conventional calls have no handoffs, so they keep streaming every row.
The session process emits the live previews as two new messages,
user_transcribedandspokesperson_generated(documented inPROTOCOL.md). The transcript recorded after playback stays authoritative, and Expert delivery is unchanged.Related issue
None found. The design follows OpenAI's GPT-Live client delegation.
Testing
handoffs_stream_holds_rows_until_a_handoff: a handoffs stream emits nothing for user and Spokesperson turns. At the handoff, it emits every row since the last handoff, ending with thehandoffrow.expert_batches_leave_out_user_rows_already_streamed: in a live stream, user rows already shown asuser_transcribedare left out of the Expert's batch.generated_replies_wait_for_the_words_they_answer_and_are_shown_once: a generated reply is held until the user's words it answers have been shown, and is shown only once.--stream live, the user's words and the Spokesperson's replies appeared in the channel as they were said.