update
Jun 3, 2026
By Teun
CTP Room puts AI coding agents in one shared chat room
A developer has built CTP Room, a shared coordination layer for humans and multiple AI coding agents. The system routes messages to the right agent, tracks file ownership, and keeps team memory across sessions.
A developer has introduced CTP Room, a shared chat room for humans and AI coding agents that is meant to replace one-on-one agent sessions with a team-style workflow. The project is pitched as a coordination and co-work layer, with the developer saying it lets people and agents work together in one room, similar to a Slack channel.
The creator described CTP Room in a Show HN post, saying the tool was built because they do not like working in separate one-on-one sessions with Claude or Codex when collaborating with a team. Instead, CTP Room creates one shared space where multiple humans and multiple agents can participate at the same time.
⚡ New to this?
This matters because a lot of AI coding tools still behave like private chatbots, even when teams need them to work together. CTP Room is trying to turn those agents into a shared workspace, where coordination, file ownership, and session memory are managed in one place.
MCP, or Model Context Protocol, is a way for AI tools to connect to outside apps and services. For non-specialists, the big idea here is simple: the project is about making AI assistants behave less like separate helpers and more like a coordinated team.
🦞 OpenClaw angle
If you build self-hosted AI automation, think about separating coordination from generation. Use a lightweight router for task assignment and reserve your larger model for the actual code or content work, so you do not burn tokens on status chatter.
Also add explicit file-locking or claim state in your agent workflows. If multiple agents can touch the same repo, keep persistent memory of decisions and ownership so one run does not overwrite another run’s work or repeat the same analysis.
One of the main features is message routing. According to the developer, the system does not broadcast every message to every agent; instead, it routes each message to the most relevant agent. For the example shared in the post, a cheaper Haiku model is used as the router.
CTP Room also includes a file-claiming system. Before editing a file, an agent claims it so other agents know who is working on it and do not overwrite each other’s changes. If another agent tries to edit the same file, it is told who currently holds the claim.
The developer said the system keeps persistent team memory as well. That includes who did what, along with decisions and rationale, so context is not lost between sessions. The goal is to preserve the working history of a team rather than forcing every new session to start from scratch.
CTP Room is designed to work with multiple agent tools. The creator said users can bring their own agent through MCP, or Model Context Protocol, including Claude Code, Codex, Cursor, and OpenCode. The system can also connect to any program over HTTP, and the developer said this multi-vendor approach is intentional.
The post also highlights a design choice meant to keep costs down. Presence and activity feeds come from deterministic hooks at zero token cost, according to the developer, while the more expensive model is only used for actual work. In other words, the system avoids spending model tokens on simple status updates and coordination signals.
The developer said they may open source the project soon and invited feedback from the Hacker News community. For now, CTP Room is being presented as an experiment in how AI coding agents can coordinate more like a team than isolated chat sessions.
