AI agents that work as a team. They ask before they act.
Give each agent a job, its own tools and its own conversation. They work together in channels on your computer and check with you before they touch anything real.
- A team, not a prompt box. Every agent has a name, a job description and a conversation of its own. Put several in a channel and they hand work to each other.
- They ask first. You choose which actions need your yes (send, pay, delete). The agent stops, shows exactly what it wants to do, and waits.
- Bring your own model. Your Claude or ChatGPT subscription, your own key, or a model that runs on your machine. Each agent can use a different one.
- It runs on your computer. Conversations, memory and files stay with you. No account is needed for the local edition.
- Open source, end to end. Apache 2.0. No hosted service is required to run any part of it.
An agent is an AI with a name, a job description, its own conversation and the tools you give it. You talk to it like a coworker. It works in the background and comes back with the result. Write one, or install one from the Store, which ships ready-made agents for finance, operations, marketing, development, support, sales, design, knowledge and people. Everything about them is editable.
Add several agents to a channel and ask. Mention one with @, or leave it to Rook: it reads each
squad's purpose and routes the request to the one that fits. Agents hand work to each other, and every
step is shown.
Set a rule per tool: allow, ask, or block. When an agent wants something that needs your yes, it posts an approval card with the exact call and waits. You can approve from the app or from Telegram.
Give an agent a Linux desktop with a browser, a terminal and files. You watch its screen live and can take control at any moment. Computers run isolated in Docker or in the cloud (E2B, Daytona, Boat, Railway, CreateOS). On your own machine, the desktop app can work in a folder of its own once you allow it; commands there are not isolated, so grant access only to what you trust.
Run an agent on a schedule or when something new shows up in a connected app. Every run leaves a history you can open.
Connect apps from the built-in catalog (Notion, Linear, Stripe, Sentry, Supabase, Vercel and more) or any remote MCP server. Each agent only gets the connections you give it, and credentials are encrypted on the server and never returned by the API.
See tokens by day and by agent, and how full your Claude or ChatGPT subscription window is when the provider reports it.
Message an agent from Telegram or Slack, or give it an inbox of its own. The reply comes back where you asked. Voice mode lets you talk with an agent: in the desktop app it listens and speaks with models on your own computer, free and offline, and a speech provider key is an option for other voices.
| Connection | Notes |
|---|---|
| Claude subscription | Sign in with your account |
| ChatGPT subscription | Sign in with your account |
| Your own API key | OpenRouter, Anthropic, OpenAI and other compatible providers |
| Local model | Any OpenAI-compatible server, such as Ollama |
Providers are configured through shared connection settings, so adding a compatible one needs no code.
Runs on your computer with its own database. No account, no server to set up.
macOS and Linux
curl -fsSL https://rook.team/install | shWindows (PowerShell)
irm https://rook.team/install.ps1 | iexOr download the app from the releases page:
corvo-mac-universal.dmg (Apple silicon and Intel), corvo-win-x64.exe, or for Linux
corvo-linux-x86_64.AppImage / corvo-linux-amd64.deb (x64) and corvo-linux-arm64.AppImage / corvo-linux-arm64.deb (ARM64).
The same command updates the app later. Details and signing notes: desktop releases.
On Linux the AppImage needs FUSE 2 (libfuse2, or libfuse2t64 on Ubuntu 24.04). Without it, run
./corvo-linux-x86_64.AppImage --appimage-extract-and-run.
Agents keep working when Rook runs on a server. You need Docker Engine, the Compose plugin, curl and OpenSSL. No clone and no Node install:
mkdir -p corvo && cd corvo &&
curl -fsSLO https://raw.githubusercontent.com/corvo-technology/rook/main/infra/compose/install-images.sh &&
bash install-images.shThe installer downloads the Compose files, creates .env with random secrets and starts Rook. It
keeps an existing .env when you run it again. Open http://127.0.0.1:5173,
create an account and connect a model.
For a public server, put HTTPS in front of port 5173 and set CORVO_HOST; the
self-hosting guide covers Caddy and host hardening.
Restricted networks and mirrors are covered
there too. To let an AI assistant do the
install for you, use SETUP_PROMPT.md. You can also
deploy on Railway.
| Local | Self-hosted | Cloud | |
|---|---|---|---|
| Where it runs | Your computer | Your server | Hosted by us |
| Account | None | Email and password | Account |
| Models | Your subscription, key or local model | Same | Same, plus credits |
| Data | On your machine | On your server | Hosted |
flowchart LR
UI["Web · Desktop · Mobile"] --> API["API"]
API --> DB[("Postgres")]
API --> Worker["Worker: runs, routines, retries"]
Worker --> Model["Model: subscription, key or local"]
Worker --> Tools["Tools: MCP, apps, approvals"]
Worker --> Computer["Agent computer: Docker or cloud"]
The frontends express intent and render state. The backend owns orchestration, authorization, validation, retries and provider translation. Models, computers, memory, voice and integrations are optional and sit behind provider-neutral interfaces. See AGENTS.md for the rules the code follows.
In the local edition everything lives in the app's data folder on your machine. Secrets are encrypted with a key kept on your machine, and content encryption is available for servers. Nothing leaves your computer except what you send to the model provider you chose.
You need Node.js 22.22.2 or newer in the 22.x line, 24.x, or 26+, pnpm 9, and Docker. Node.js 23.x and 25.x are not supported.
git clone https://github.com/corvo-technology/rook.git
cd corvo
cp .env.example .env # then set the secrets listed inside
pnpm install
pnpm sandbox:build
pnpm devpnpm dev starts Postgres when needed, applies migrations and prints the URL. Required settings,
database roles, ports, providers and troubleshooting are in the self-hosting guide.
apps/ web, api, worker, desktop, mobile and the public website
packages/ domain, contracts, persistence, adapters, UI and test tooling
infra/ local services and computer images
docs/ architecture, operations and release guides
pnpm lint
pnpm check
pnpm testLaunch the desktop app against the dev stack with pnpm --filter @corvo/desktop dev and choose
Existing instance with http://127.0.0.1:5173. See CONTRIBUTING.md for the
workflow and the full test matrix.
The app is available in English, Português (Brasil), Español, Français, Deutsch, 日本語 and 简体中文 under Settings → Language.
- Self-hosting and secrets
- Sandbox providers
- Computer runtime and isolation
- Desktop releases and manual release
- Mobile releases
- Performance testing
Contributions are welcome. Read CONTRIBUTING.md before opening a pull request. For security issues, follow SECURITY.md instead of filing a public issue.
Apache License 2.0. See THIRD_PARTY_NOTICES.md for bundled software.






