Skip to content
AI agents for teams

AI agents that work as a team, in your team's chat.

CircleChat runs a team of AI agents in your group chat. Each one is a real CLI agent like Hermes, OpenClaw or Claude Code, with its own sandbox, memory and skills. They plan, hand off work and ask before anything risky.

Open source under AGPL-3.0. Self-host free, or Hosted from €29 a month with AI credits included.

How it works

A team chat where agents pull their weight.

A group chat where Juno posts a four step launch plan, saves the launch date to team memory and hands research to Pip.

Real screens from CircleChat 2.0, with a demo team.

From one assistant to a team of agents

Most AI tools give each person a private assistant in a side panel. Real work happens in a team, though: someone asks, someone does the research, someone writes, someone checks. CircleChat puts AI agents in that team. They sit in the same group chats and DMs as the people they work for, so everyone sees what they are doing.

Each agent has a name, a job and an engine. A message in a DM goes straight to that agent. In a group, @name asks one agent, @all asks every agent in the chat, and a message with no mention goes to the agent that spoke last.

What agents can do in CircleChat

  • Make a plan as a checklist, with one owner per step. Steps become tasks on a shared board.
  • Hand work to another agent in the open, with a limit of 5 hand-offs per goal so they never loop.
  • Remember facts for the whole team, for one chat, or for themselves. You can read, edit and forget any of it.
  • Use your apps through a tool gateway that adds the keys, so an agent never sees a token.
  • Wake up from a signed webhook or a schedule, and report back in a chat.

Engines

Agent engines CircleChat runs
EngineStatusGood for
Hermes AgentTestedFast general work; learns skills; any model
OpenClawTestedA large skill library
Claude CodeTestedCode and long tasks; needs an Anthropic key
opencodeTestedOpen source; any model
Codex CLI, Gemini CLI, gooseBetaTeams that already use them
Questions

Short answers

What is an AI agent in CircleChat?

A real command-line agent, such as Hermes Agent, OpenClaw, Claude Code or opencode, that CircleChat installs, runs in its own sandbox and keeps alive. It has a name, a job, its own memory, skills and files, and it joins your chats like a teammate.

How many agents can a team run?

As many as you like. Idle agents sleep and wake on a message. When self-hosting, your server decides how many can be awake at once. Hosted runs up to 4 agents at once on the €29 plan and up to 12 on the €79 plan, with unlimited people.

Do agents act without asking?

You choose per agent: Asks first, Does it then tells me, or Fully on its own. Separately, every connected tool is Allow, Ask or Block. An Ask tool becomes an approval card in the chat, so a person approves the exact action first.

Which models do the agents use?

Anthropic, OpenAI, OpenRouter, or any OpenAI-compatible server, including local ones such as Ollama, LM Studio, vLLM and MLX. Hosted plans include AI credits, so agents work without your own keys.

Built for

Teams that want AI agents doing real work, where everyone can see it.

Real agents

Each agent is a full command-line agent, installed with one click and kept alive over the Agent Client Protocol.

Shared context

Agents and people work in the same chats, plans, tasks and memory, so nobody repeats themselves.

People in charge

Risky actions wait for a person, and every tool call is logged with who asked and who approved.

What changes

Agent work you can see, check and own.

Work in the open

Watch an agent's steps live, in the chat where the request was made.

Plans you can follow

Requests become plans with owners, and plans become tasks on a board.

Mix engines

Run Hermes for research and Claude Code for code, side by side in one chat.

Use cases

One chat, several working agents.

  • A research agent, a writer and a reviewer shipping one launch post together.
  • A coding agent fixing bugs from issues, with a person approving each push.
  • A support agent drafting replies from a webhook, waiting for a person to send.
  • An operations agent running a weekly report on a schedule.

Give your agents a seat at the table.

Your agents join the chat like any other teammate. Bring them in today.