Team chat turns one chat message into a small multi-agent run. Mention two or more agents — @claude @codex, or even @claude @claude — and Vulgr starts a background pipeline instead of a single reply.
Starting a team
- Type more than one
@agentin the same message. Two mentions are enough to switch into team mode. - The same agent can be added twice. Each copy thinks independently and questions the other.
- Every agent gets its own model: pick one per agent in the chat header.
- Fast mode runs team rounds on faster models so you aren’t left waiting.
The pipeline
Under the hood a thread-based pipeline (teamChat.ts, threads/) runs three stages:
- Discussion — the agents share one thread. They split the work, propose, object and change their minds when they should.
- Verification — the outcome isn’t taken on trust. It is checked with your project’s own tests, run locally.
- Synthesis — the thread is folded into one summary: what was decided, what changed and what the tests said.
Discussion drawer
The answer you see is the synthesis. Everything that led to it — every argument and tool step — sits in a collapsible discussion drawer below it, so the chat stays readable but nothing is hidden.
Approvals
When an agent in the team wants to run a command, the approval broker (approvalBroker.ts) asks you: allow, always allow in this chat or deny. You can also set the default to ask, auto-allow or always deny.