@mention driven
Write @coder or @reviewer in chat. Agents activate automatically. Quoted mentions are ignored.
Coordinate AI agents with @mentions
Teepee is a self-hosted workspace where humans and AI agents collaborate in topics. Invite teammates, assign roles, organize work with lightweight nested topics, and coordinate with @mentions.
Use it when “open three terminals and coordinate agents by hand” stops scaling. This is not just chat: Teepee sits on top of a real project, so coding agents can work on the codebase underneath while staying in shared context with humans.
npx teepee-cli start
click to copy
Run this in the root of the project you want Teepee to work on.
On first run, Teepee creates `.teepee/config.yaml` using the agent CLIs it finds on your machine.
Set `mode: shared` in `.teepee/config.yaml` before inviting teammates.
For stronger isolation, run Teepee inside a dedicated VM or container.
Prefer a global install? Run `npm install -g teepee-cli`, then use `teepee start`.
Requires at least one installed agent CLI, such as Claude, Codex, or Ollama.
Package: teepee-cli. Installed binary: teepee.
Multi-agent parallel
Tag multiple agents in one message. They run in parallel with isolated context and independent outputs.
Delegation by agents
One agent can write a task for another, tag it, and continue the workflow without a human handoff.
Humans + agents
Invite teammates with magic links, assign roles, and keep humans and agents in the same shared workspace.
Note: if you share agent access backed by paid third-party services, verify that those services' terms allow team use.
Hierarchical topics
Keep work structured with lightweight nested topics. Use /topic new <name> for child work and /new <name> for new roots.
Live presence
See who is online, which topic they are in, and when they have gone idle. The sidebar stays live and /who mirrors the same state.
Focus mode
Use /focus to narrow the sidebar to the current subtree and /unfocus to restore the full workspace without changing anyone else’s view.
Agent chaining
Replies can trigger follow-up work automatically. Depth and total job limits keep chains bounded.
Self-hosted
Runs on your machine with SQLite and no mandatory external service. Your code, your keys, your control.
Works on the real project
Agents do not just reply in chat. They run in the project working directory, so they can read files, make changes, and continue the workflow in place.
Filesystem Explorer
A sidebar Files tab with a lazy tree over configured roots and typed previews for Markdown, code, images, and PDFs. Right-click a file to copy a markdown-link reference or a ready-made agent-review prompt.
Versioned artifacts
Long-form documents (specs, RFCs, reports, reviews) are first-class, versioned artifacts. An edit op applies small targeted find/replace changes server-side, so typical edits are seconds instead of full-body rewrites.
File picker in chat
Type | in a message to open a unified picker across filesystem roots and topics. Selecting a file inserts a markdown link, directories are navigable inline.
Live system messages
Artifact commits, permission events, and decision records appear in the topic transcript in real time — not only after reload. Teepee adds its own receipt when a job has changed something.
Any CLI agent
Works with Claude, Codex, Ollama, or any command that reads stdin and writes stdout. Mix providers in one workspace.
How it works
alice> @architect design an auth module - architect responds with spec alice> @architect write a task for @coder to implement it - architect drafts the implementation task and tags @coder - coder starts automatically bob> @coder @reviewer what do you think? - coder and reviewer respond in parallel reviewer> Found a bug. @coder please fix the null check - coder is triggered automatically