Tico

Tico is a source-available operating system for a team of humans and bots, licensed under PolyForm Perimeter 1.0.1. Humans file work, answer bots and approve actions in a web app; bots pick the work up, run it with the team's own model subscriptions, and report back. It is built for teams from a handful of bots up. A small pilot is what has been measured (sizing).

• A small server holds the team. One Docker install (the server, HTTPS through Caddy or a Cloudflare Tunnel, Litestream backups, an optional updater) keeps tasks, chats, automations, files and humans in SQLite. The server never runs a bot or a model CLI; when decisions (or Slack routing) are on, it sends the text of each question to the decision model: the Tico AI Gateway on Tico HQ with a gateway key, then your own TypeSafe key if you set one, then OpenAI, Anthropic, Gemini, xAI or OpenRouter with a key stored on the server.
• Computers run the bots. A Mac (the native runner) or any Linux or cloud computer (the tico-runner image) joins with a one-time code, claims work over HTTPS, and runs each bot's run in that bot's own git repository. Model sign-ins stay on the computer; granted bot credentials can also come from Tico's encrypted vault.
• Harnesses are installed per computer, on demand. Codex, Claude Code, Gemini CLI, Grok, or a multi-provider catch-all; each bot picks one. Humans sign in with built-in Google or Microsoft sign-in, and GitHub access comes from a per-team GitHub App scoped to each bot's own repository.

Quick start

On your own computer (a Mac with Docker Desktop, or Linux), with no domain and no sign-in setup, run the installer with --local:

curl -fsSL https://github.com/ticoteam/tico/releases/latest/download/install.sh -o install.sh && sh install.sh --local --owner-email [email protected]

Docker Desktop must be open and running on a Mac. Tico runs at http://127.0.0.1:8765; open the sign-in link the installer prints. Then get one first result: Content Owner drafts a short post about a made-up product, you review it, and you close the task. Sign in to each model subscription you chose (a ChatGPT sign-in for Codex, a Claude sign-in for Claude Code), or use an API key. One signed-in model is enough for the first result; Health lists the others until you sign in. You need no GitHub account or outside Tool. Demo mode does not run bots.

1. In Finish setup, choose an AI provider, name your team and pick Marketing with Content Owner (content) checked. With no provider yet, write one bet and one line under What we won't do; any short sentence works. Ten built-in bots are created: Assistant, BotOps, Librarian, Strategy Owner, General Manager, Finance Owner, Rules Owner, Design Owner, Updates Owner and Automations Owner.
2. Use Add computer to join this computer, run the command shown, then Create my team.
3. Open Settings > Computers and Sign in beside Codex or Claude Code; finish the provider's browser flow. Other harnesses and API credentials are covered in Harnesses. Choose the bot's model in Settings > Bots.
4. Before any real work, open Settings > Health: Models should read "Codex is signed in." (or Claude Code). If not, follow first-result recovery. With no spending limit set there is no cap; a limit pauses new work and lets a run in progress finish (Usage).
5. On the team chart, open each seat and press Hire on its template, in any order. A seat hired before its manager reports to the nearest hired manager above it until that manager is hired, then moves under it. On Content Owner's page, press Start onboarding. Use the sample answers to give it readers, publishing plans, topics, voice examples and privacy rules; review its first draft in chat.
6. On Tasks, press + (New task) and assign it the short draft from Your first result. The bot marks it Done with the draft file on the task. Read the draft, then close the task yourself: the requester closes it, not the bot.

Success is a draft file on the task, read by a person, and a task you closed. The steps, and what to try next (meeting notes with Project Coordinator), are in Your first result.

You can choose providers later in Settings > Computers > Models; bots wait with “Add an AI provider” until you do. Add a domain and sign-in later (Install).

For a team, on a Linux server (about 2 GB, with a domain pointed at it), run the installer with no flags:

curl -fsSL https://github.com/ticoteam/tico/releases/latest/download/install.sh -o install.sh && sh install.sh

It installs Docker if needed, downloads that release's compose bundle (checksum verified) and walks you through the setup wizard (python3 -m setup): a domain name to a signed-in Tico, with each step checked. Then open the app, complete Finish setup, and use Settings > Computers > Add computer to join a Mac or a Linux computer. What you need first, where to get a server, and what you will see are in docs/install.md.

Just looking? docker run --rm -p 127.0.0.1:8765:8765 ghcr.io/ticoteam/tico:latest demo opens a fictional team on localhost with no setup (docs/demo.md).

Documentation

Start here · Use Tico · Developer and computer operator · Glossary

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗