Calls: decide it out loud, and the agents can quote you
Huddles transcribed with the speaker on every line, summarized with action items, and readable by any session.
A lot of engineering decisions are made on a call, and coding agents never hear them. Codecast huddles are built into the workspace (channels, DMs and each session's own room) and transcribe while they run, with the speaker's name on every line. When a huddle ends it gets a title, a summary and action items, and any agent can read it, quote the exact words, and turn what was agreed into tasks.
Starting a huddle
The headphones button in a channel's header starts a huddle and buzzes everyone in it; in a DM it is push to talk. In a session, the same button starts a huddle with the agent, so one person is enough for a call. The phone can start or join one from any channel or session. Expand the call opens the stage with video, screen share and live captions under each speaker's name, and a transcribing pill shows whether words are being captured. Calls are a team feature an admin turns on in Settings.
Every line knows who said it
Most meeting tools record one mixed stream and guess speakers afterwards. Codecast never mixes: one app in the room acts as scribe and sends each participant's own audio track to speech recognition separately, so attribution is structural rather than guessed. If the scribe's laptop closes, another participant's app takes over without writing a word twice.
The digest lands where the huddle was held
A huddle in a channel or DM posts a digest there when it ends: "6 min huddle with Maya, Theo and Priya", a summary and the action items, with a link to the transcript. A huddle in a session's own room wakes that session's agent instead, with a short summary; the full transcript stays on the server until the agent asks for it, rather than flooding its context.
Every call also lives on the Calls page, live ones first. A call's page has the transcript, the summary ("So far" while it is live) and the action items; select a few turns to copy a link to them or Send to agent. A call has a short id like a task, and every line has its own reference. cl-42 in a message renders as a pill with the call's title, and cl-42:3-4 opens the call at those lines. So when an agent says "Theo agreed to take the retries", it links to the line where Theo said it.
Action items become tasks that quote the call
The generated action items are a draft, so nothing files them blindly. Ask an agent to turn a call into tasks and it reads the summary, checks every item against the transcript (who committed, to what, by when, and whether anyone pushed back later), drops what was only floated and adds what the summary missed. Each surviving item becomes a task assigned to the person who took it on, with their words quoted in the description and marked as decided in a meeting, so it reaches the human task board on its own.
Agents on the call
Add to the call brings in a teammate, a role or an agent session, which then "hears the room and answers here". Words reach the agent each time the room goes quiet; a line that names it goes through at once, even mid-turn. When it needs to concentrate it can put the room on hold for up to half an hour and catch up after. It answers in the huddle's chat, and its face carries an Agent tag.
Recording and screens
Recording is separate from transcription and never silent: only Record this huddle starts it, everyone (later joiners included) is told, a REC mark stays up, and it stops when no teammate is left. Screen shares are kept at full resolution, and an agent can pull the frame from the moment a line was said, so it sees the code that was on screen when someone said "this function here".





