Your team of AI agents, on your Mac, Windows or Linux. Agents that talk in channels, write code, review each other and use your tools, on the Claude Code, Codex, OpenCode or Pi you already have. Anything that touches the real world waits for your yes.
TypeScript agent-teamagentic-aiai-agentsai-assistantclaude-codecodexcursorelectrongrokgrok-bot-alternativelinuxlocal-firstmacosmcpmulti-agentopenclaw-alternativeopencodewindows
Security
Tucano runs AI agents that can read your tools, run commands and click in a browser, so its permission model matters. If you find a way around it, please tell us privately.
Reporting a vulnerability
Use GitHub's private vulnerability reporting. Please don't open a public issue. Include what an attacker could do, the steps to reproduce and the version you tested. You'll get an answer within a few days.
Especially interesting:
- an agent using a tool it wasn't granted, or a write tool with read-only access;
- a sensitive action running without its approval, or an approval that runs something other than what was shown;
- an agent reaching files outside its computer and the folders lent to it;
- content from a web page, an issue or an email making an agent act on its own (prompt injection that crosses a permission boundary);
- secrets (tokens, connector logins) leaving the vault-encrypted store (macOS Keychain, Linux keyring) or reaching a model.
How Tucano keeps agents in bounds
- Each reply runs the harness CLI isolated from your personal configuration, with no built-in tools. Its only tools come from Tucano's local MCP gateway, through a secret per-turn URL that lists just that agent's tools; any other call is refused.
- Write actions that matter (merge, publish, send, delete) become an approval tied to the exact call and arguments. Approving runs exactly that call, after checking the grant again.
- Connector secrets are encrypted with a key in the macOS Keychain (on Linux, the system keyring: GNOME Keyring or KWallet) before they reach the database. Without a keyring, Tucano refuses to store them.
- An agent's computer is a Docker container with only its own folder and the folders you lend it, or, if you choose so, your own computer.
- Everything stays on your computer: conversations, memory, logins. There is no Tucano server.