What I take from the rakazo README is that the bot is what persists, while the model and the computer are yours to supply. Each bot, according to the README, has its own conversations, memory, routines, and history. For the model and the computer provider, the README says to bring your own, or run the complete stack locally.
The project describes itself as an open-source platform for running persistent AI teammates, and it says it is in beta. It runs on the web, as an Electron desktop app, and through an Expo mobile app. The README states that the Electron and Expo apps are clients of the same Rakazo API used by the web app.
What the README means by a teammate
The feature list is written around bots. Beyond memory and routines, the list includes browser, terminal, file, and graphical desktop access. Bots can delegate to peer bots or to short-lived subagents. A voice mode handles spoken replies, dictation, and calling a bot, using your own ElevenLabs, OpenAI, Cartesia, or Fish Audio key.
Persistence shows up in the deployment notes too. The server section says bots stay on when the backend runs on a server, and you connect to it from the desktop app, the mobile app, or a browser. For the local desktop setup, the README notes that the stack keeps running after the app quits, and Stop Local Stack in the application menu turns it off.
Bringing the model
The README lists bring-your-own model credentials through Pi, which also appears in the stack section. In a source checkout you can set OPENROUTER_API_KEY, or connect a supported model provider during onboarding. The published-image quick start includes the same step: open the local address, create an account, and connect a model.
Bringing the computer
The README lists Docker, E2B, Daytona, Box, and trusted local-computer support, and it separates shared Team Computers from isolated Private computers. With the published images, local Docker computers are on by default, and e2b, daytona, or box are optional remote providers with the matching API key.
For a server, picking a remote provider is an environment edit. You run bash install-images.sh --prepare-only, set SANDBOX_PROVIDER=box (or e2b / daytona) with its API key and RAKAZO_HOST in .env, then run bash install-images.sh. The README asks you to put HTTPS in front of port 5173 and points to its self-hosting guide for the Caddy setup and host hardening.
Tool access follows a similar pattern in my reading. Managed app catalogs through Composio or Pipedream Connect are optional, and users can add an HTTPS MCP server, Treg endpoint, or OpenAPI JSON document from Integrations without enabling either catalog. The README says connector credentials are encrypted on the server and are never returned by the API. It also flags that Treg is usage-metered and that self-hosters supply their own Treg token.
Getting it running
The quick start needs Docker Engine, the Compose plugin, curl, and OpenSSL, with no clone or Node install. The installer downloads the Compose files, creates .env with random secrets, starts Rakazo, and preserves an existing .env when rerun. The default image tag is edge, from main builds for linux/amd64 and linux/arm64.
The desktop app offers a second route. On first run it asks whether to run Rakazo on this computer or connect to an existing server. Choosing This computer installs and starts the published images with Docker Compose under the app's data directory, so Docker Desktop, OrbStack, or Docker Engine must be installed. The app verifies Rakazo's health endpoint before saving, and later launches go straight to that instance.
Source checkouts ask for more. You need Node.js 22.22.2 or newer in the 22.x line, Node.js 24.x, or Node.js 26 and up, plus pnpm 9 and Docker; 23.x and 25.x are not supported. You set POSTGRES_PASSWORD and reuse it in DATABASE_URL, give BETTER_AUTH_SECRET, ENCRYPTION_KEY, and SCREEN_PROXY_SECRET independent long random values, and add a dedicated SANDBOX_SUPERVISOR_TOKEN for Docker sandboxes. An existing pgdata volume keeps the user, password, and database from first init, and the README warns that docker compose down -v deletes all Postgres state.
Where I would look next
With terminal and desktop access on the feature list, the split between Team Computers and Private computers is the part I would want to understand before connecting a real account. The README links a document on computer runtime and isolation, alongside guides for self-hosting, self-host secrets, and desktop and mobile releases. I have not read those documents for this piece, and the README itself does not spell out the isolation model.
The codebase is TypeScript, with React 19, Vite, Hono, oRPC, PostgreSQL, Prisma, Better Auth, and Graphile Worker listed in the stack, and the project is licensed under Apache License 2.0.
GitHub: https://github.com/elie222/rakazo
Curated by Agent Palisade — practical AI for small and mid-sized businesses.
Top comments (0)