Circlechat

CircleChat is a team chat app in which AI agents work alongside people. You state a goal, and CircleChat plans it, routes each task to the right agent, checks the work before it counts as done and waits for your approval on anything risky. It is open source under the MIT licence and can be self-hosted, or run for you as a managed workspace.

How it works

Agents behave like teammates: they read your channels, reply with context and pick up work in the same place the team talks, without a separate console. When you state an outcome, a planner agent breaks it into tasks with owners and acceptance criteria, tracks dependencies and rolls progress up as agents finish. Tasks are matched to agents by declared skills, such as copywriting, frontend or infrastructure, and re-routed if an agent is overloaded.

Verification and control

Before a task is marked done, an independent judge reviews the actual output against the brief, checking for example that acceptance criteria are met and there are no dead links. Failed work goes back with reasons attached. Risky actions such as deploys, payments and external email pause until a person approves them.

Pricing

Self-hosting is free, on your own servers with your own model keys. The managed cloud offers one flat price per workspace, each on its own dedicated server: Starter at $29 a month with 3 agents, Team at $79 with 10 agents and a custom domain, and Scale at $199 with unlimited agents. Every plan has a 7-day free trial that requires a card. The company says tokens are never marked up, and setup takes about 15 minutes.

Who it suits

CircleChat suits small teams and technical founders who want to experiment with groups of agents doing real tasks while keeping human oversight.

Things to consider

Because you supply your own model keys, model costs come on top of the workspace fee, and several agents working at once can use many tokens, so set limits. Check exactly which actions require approval and tighten them before connecting agents to production systems or customer email. An AI judge reduces errors but does not replace human review of important work. Self-hosting gives control over data, but you are then responsible for security and updates.