Hi HN. I honestyle DO NOT like one on one sessions with my claude/codex when working with my team.
So I built something that connects us and our agents in the same room.
CTP Room is my attempt at a coordination/co-work layer, one shared chat room (think Slack channel) where humans and several agents work together ( I took Garry Tans gstack and formed my own team around it when I work solo ).
What it does:
- Routes each message to the right agent instead of broadcasting (For this case, a cheap Haiku router picks one).
- File claims: an agent claims a file before editing; another is told who holds it instead of
clobbering it.
- Persistent team memory: who did what, decisions + rationale, so context survives sessions.
- Bring your own agent over MCP (Claude Code, Codex, Cursor, OpenCode) or any program over HTTP.
Multi-vendor on purpose.
- Presence + activity feed come from deterministic hooks at zero tokens; the expensive model is only
spent on real work.
Happy to go deeper on this subject, I think I will open source it soon, wondering what you think about this?
What it does: - Routes each message to the right agent instead of broadcasting (For this case, a cheap Haiku router picks one). - File claims: an agent claims a file before editing; another is told who holds it instead of clobbering it. - Persistent team memory: who did what, decisions + rationale, so context survives sessions. - Bring your own agent over MCP (Claude Code, Codex, Cursor, OpenCode) or any program over HTTP. Multi-vendor on purpose. - Presence + activity feed come from deterministic hooks at zero tokens; the expensive model is only spent on real work.
Happy to go deeper on this subject, I think I will open source it soon, wondering what you think about this?