Skip to content

Questions and Actions

While the bot is in a call it picks out moments: things said that something can be done with. Each moment offers actions, and each action asked is a run.

Questions

A question asked in the call (not one put to someone in it, like can you send that over?) shows up on the meeting's Questions tab while the call is on. Questions asked close together read as one card. After the call, where the stack has a model, they are cleaned up and grouped by topic. Look over the call again (or meetings look-over) reads an old call over again.

Actions

Action What it does
Copy for agent Copies a prompt (the last lines of the call, then the question) to paste into Claude Code, Codex or anything else
Ask agent A coding agent on a runner answers it from your repository
Ask Illiana Illiana answers it in the chat

A meeting in Overseer's drawer while the bot is in the call: a question asked by name, its actions, and Illiana's answer under it

Each run shows under its question: waiting, answering, then answered or failed. A run claimed and never finished fails after half an hour.

To ask an action of every question as it is heard, set MEETINGS_AUTO_ACTIONS (Overseer > Meetings > Settings), for example question=agent.

The runner

The coding agent runs on your machine, not in the app. The runner asks the app for waiting runs, answers each in a fresh worktree of your repository, and posts back what it printed.

 "Ask agent"  --->  run: asked
                    POST /runs/claim   <---  runner claims the oldest
                                             fresh worktree of --repo
                                             runs --command
                    POST /runs/{id}/finish <-- posts what it printed
                    run: done | failed
my-app meetings runner-token create laptop
export MEETINGS_RUNNER_TOKEN=mrt_...
my-app meetings runner \
  --url https://my-app.example.com \
  --repo ~/code/my-service \
  --command 'claude -p {prompt} --permission-mode plan'

Any headless agent works (codex exec {prompt}, aider --message {prompt}). A runner answers one run at a time, oldest first; start more to answer more at once. In the dev stack a meetings-runner service runs one for you, configured in .env.meetings-runner.

Your own moments and actions

Register them in a meeting_moments.py beside any service's service.py:

from app.services.meetings.service import Action, register_action, register_detector

register_detector("decision", lambda heard: heard.text if "we decided" in heard.text.lower() else None, title="Decisions")
register_action(Action("ticket", "File a ticket", frozenset({"question"}), "runner"))

An action runs in the browser (its prompt copied), the app, or on a runner started with --action ticket.

Was this page helpful?