Image: GitHub
Why it mattersTeams already pay for Slack and for AI tools. A single app for both puts agent work in the same thread as human work.
agentes.chat is a self-hosted team chat where human teammates and AI agents share the same channels and direct messages. The repository went up on GitHub on 9 November. It has 1,054 stars, 158 forks and 7,952 commits on the main branch at the time of writing.
The pitch is straight from the README. "Open-source, self-hosted Slack alternative where your team and AI agents work together in channels and DMs. Agents run on Claude, ChatGPT/Codex, Gemini or Grok, with MCP tools, their own computer and browser, routines, memory and approvals."
What it is
Agents sit in channels as members. A user can mention one the same way as a teammate. The agent reads the thread, picks a tool, and writes back. Tools come from the Model Context Protocol. The README names Linear, Sentry, Notion and Stripe among the built-in connectors.
Each agent has its own computer and headless browser, per the README. Work it finishes outside chat shows back up as a reply with a link. The agent also has memory that holds across sessions. An admin can gate a sensitive action behind an approval. A merge or a deployment then waits for a human reply before it runs.
Shape of the product
The project includes a web app, a desktop app for Mac, Windows and Linux, and phone apps. Channels, threads, direct messages, a wiki with version history and task boards are in the README feature list. Multi-language support covers 14 languages.
Setup is one Docker Compose file, per the README, with PostgreSQL as the store. A no-container path is also listed for small teams. The code is TypeScript. The licence is AGPL-3.0 for the server and MIT for the setup CLI. The agentes-chat GitHub organisation credits Codari Global, LLC in the NOTICE file.
Why this shape, now
Teams already pay for Slack or Microsoft Teams. Many of the same teams also pay for one or more AI assistants. The usual pattern today is a separate chat tab for each AI. The team chat holds a separate thread where somebody pastes the AI's reply. A tool that puts both in one channel removes the paste step.
Building the agent as a channel member also fixes a quieter problem. A model that works inside a thread sees the context around the question: the previous replies, the files that were shared, the people who are on the channel. A model called from a browser tab sees none of that.
Star growth over three weeks tells less than three months of production use. The project is new. Features marked beta in the README include memory, the mobile app and some of the MCP connectors. A team that would depend on it today should read the issue list first. Keep a fallback open before moving a real workflow.
A team that already runs its own PostgreSQL and its own chat can try the full loop for free. The loop is: human asks, agent does, human reviews, all in one channel. For the rest of the market, the signal is that chat and agents are collapsing into one product. The two names for that product today are Slack with Agentforce and this. The second one is open source.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
