teepee.org

What It Is

Teepee is a self-hosted workspace that lets human teams and AI agents collaborate in real time around real code projects. It is not just a chat tool: it runs on top of a project root directory, where agents can read files, modify code, and preserve context inside shared topics. You can start it with npx teepee-cli start or by installing teepee-cli globally. On first launch, it generates a configuration based on the agent CLIs already installed on your machine.

Core Capabilities

Teepee is built around @mention-driven multi-agent coordination. In chat, users can write @coder or @reviewer to trigger a specific agent, or tag multiple agents at once to run them in parallel, each with isolated context and separate output. Going further, one agent can write a task and @mention another agent, enabling automatic delegation and chained workflows. The system constrains these automated chains through depth and total-task limits.

For project management, it offers lightweight nested topics, real-time presence, focus mode, a filesystem browser, an in-chat file picker, and versioned artifacts for specs, RFCs, reports, and reviews. These design choices make it feel more like a development collaboration console for the AI era than a standard AI chat window.

Models, Integrations, and Privacy

Teepee does not provide models itself. It requires at least one CLI agent to be installed, such as Claude, Codex, or Ollama, and it also supports arbitrary stdin/stdout commands, so different providers can be mixed. On privacy, it is self-hosted, uses SQLite, requires no mandatory external service, and leaves code and secrets under the user’s control. If you use third-party paid model services and share them with a team, you should verify their team-use terms yourself. The source text also recommends running it in a dedicated VM or container for stronger isolation.

Pricing and Access from China

The collected information does not disclose Teepee’s pricing, free quota, or commercial support. The tool itself can be launched via npm/npx, but underlying services such as Claude or Codex may incur costs. Access from China cannot be determined from the text alone. If it depends on overseas model CLIs, networking and payment may be constrained by the relevant services. Alternatives include Cursor, Claude Code, Codex CLI, Continue, Aider, and local Ollama workflows.

Pros, Cons, and Best Fit

Its strengths are self-hosting, strong user control, the ability to put multiple AI coding agents into the same project context, and support for parallel execution, delegation, file references, and artifact versioning. The downsides are a higher setup barrier than SaaS tools, dependence on local CLIs and the quality of underlying models, and undisclosed information around Chinese-language support, enterprise security, and commercial support. It is best suited to developers and small engineering teams that already use AI coding agents but are starting to find “manually coordinating multiple terminals” hard to scale.