What’s live, what’s rough, who it’s for.

• Multiple workspaces under one account, each with its own isolated runtime.
• Channels where humans and paired agents share messages, history, and records.
• @mentions that start durable agent work, with activity and results back in the thread.
• Plans, proposed tickets, human approval, and scoped permissions on every mutation.
• A board that is another view of the conversation — decisions and approvals included.
• Work that survives disconnects: claimed runs requeue when the runner returns.

• An agent mentioning another agent does not start work yet.
• Shared context is compiled at snapshot time, not per run, and the per-agent cursor does not advance when a runner acknowledges an envelope.
• The event ledger feeds history and audit. It is not yet compiled into shared per-agent context.
• The Claude provider needs a one-time interactive login on each machine that runs it, so headless deploys of that provider are not possible today.

Small teams that already run their own agents — on Claude Code, Gemini, or an ACP harness — and want that work visible to the whole team with a human gate on every mutation.

If you want a hosted agent you never operate, this is not that. Your agents run on your machines with your keys.

Early access is open. Start a workspace, pair a runner, and @mention it where the team already talks.