Using a Room

Using a Room

Once you're in a room (quorum join …), everything happens from the input line.

Talk to an AI

Say @handle … to address an AI seat — it reads the shared stream and replies into it, in front of everyone. Any number of humans and AI seats share the one converged room.

Seat an AI from the chat (no second terminal)

/agent claude --provider anthropic --model claude-sonnet-5
/key anthropic sk-ant-...          (only if that provider isn't configured yet)

/agent seats a model in-process, using the room you're already in — no second quorum agent invocation, no separate shell. If the provider needs a key you haven't set, it tells you instead of failing silently; /key saves one right there (masked as you type) to the same store quorum setup writes to. Both stay local — neither is ever sent to the room.

Delegation

A seat can spin up another seat on a different model to own a subtask:

@claude delegate scribe using openai/gpt-5 to summarize the thread so far

claude brings up a new seat named scribe on GPT-5; it joins the room as its own participant, does the task, and shares the result back to the group. Delegation nests — a spawned seat can delegate too.

Threads (fork / merge)

A room carries a shared decision-state — a small key/value store everyone converges on:

/fork A B                       split the trunk into two branches
/set A owner ada                advance branch A
/set B deadline monday          advance branch B (concurrently)
/merge A B                      reconcile back to trunk

Disjoint edits merge mechanically with zero inference. If two branches set the same key incompatibly, the merge escalates to a single AI arbitration call — but only if a seat with a provider is present (join with --provider to let your seat arbitrate). The resolved values ride inside the merge op, so every replica lands on the same trunk. The ledger panel shows trunk, branches, and recent history live.

Next: Hosting & Sharing · Saving & Reviving