Rakazo AI agent: self-hosted persistent teammates
/rakazo
Rakazo is a good self-hosted AI agent platform if you want persistent bots, selectable models, and real computer access rather than a chat-only wrapper. The breadth is impressive, but beta status, PostgreSQL, secrets, sandbox providers, and production hardening make it an operator-led choice rather than a casual install. For a technical team willing to own the stack, I think it is worth the operational work.
What Rakazo is
Rakazo is an open-source, self-hosted AI assistant for persistent teammates. It runs on the web, Electron, iPhone, and Android, connects to a user-selected model through Pi, and gives bots isolated or shared computers with browser, terminal, file, and desktop access. Rakazo's model agnostic approach can use local infrastructure or several remote sandbox providers.
Persistent AI teammates, on your infrastructure
Persistent AI teammates need conversation history, routines, memory, and controlled access to a computer without tying each task to a fresh chat window. Rakazo fills that gap by keeping bots alive on infrastructure you operate, giving them shared team computers or isolated private computers, and connecting those bots to models and external tools you choose.
Agent, computer, and integration features
- โRakazo keeps separate conversations, persistent memory, routines, and history for each persistent AI agent.
- โBring-your-own-model AI agent access goes through Pi, allowing operators to connect their own model provider credentials.
- โComputer sandbox options include Docker, E2B, Daytona, CreateOS, Boat, and trusted local machines.
- โShared Team Computers and isolated Private computers give bots different collaboration and containment boundaries.
- โRakazo supports multi-agent delegation through peer bots and short-lived subagents.
- โVoice mode can speak replies, accept dictation, and place calls using ElevenLabs, OpenAI, Cartesia, Fish Audio, or voices installed on a Mac server.
- โApp integrations can come from Composio, Pipedream Connect, user-installed Treg endpoints, remote MCP servers, and OpenAPI documents.
Ways to use Rakazo
- โขRakazo can serve as a private research assistant whose persistent memory survives across sessions.
- โขDevelopers can give bots terminal, file, browser, and graphical desktop access for coding and operational tasks.
- โขTeams can build persistent routines around shared computers while keeping sensitive work in isolated private environments.
- โขRakazo supports multi-agent workflows in which one bot delegates a bounded task to a peer or short-lived subagent.
- โขVoice-enabled Rakazo clients support spoken conversations, dictation, and bot calls from supported devices.
- โขRakazo can connect bots to external services through managed catalogs, MCP, Treg, or OpenAPI sources.
Self-hosting with published images
Rakazo's published-image path avoids a source checkout and Node installation. You need Docker Engine, the Compose plugin, curl, and OpenSSL. Run: mkdir -p rakazo && cd rakazo && curl -fsSLO https://raw.githubusercontent.com/elie222/rakazo/main/infra/compose/install-images.sh && bash install-images.sh The installer downloads the Compose files, creates an `.env` with random secrets, and starts Rakazo while preserving an existing `.env` on later runs. 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 follow the production deployment guide.
Why self-host Rakazo
- โRakazo gives operators control over model credentials, conversation storage, computer environments, and deployment location.
- โPersistent memory, routines, history, and delegated subagents make it more useful for ongoing work than a stateless chat client.
- โOne API supports browser, desktop, and mobile clients, avoiding the need to maintain separate assistant frontends.
- โRakazo supports both shared team computers and isolated private computers, so teams can choose the boundary a bot receives.
- โConnector credentials are encrypted on the server and are not returned by the API, which provides a sensible self-hosting boundary.
Beta status and operational requirements
- โณIt is beta. The README explicitly labels Rakazo as beta software, so production operators must accept a changing product.
- โณA public deployment needs PostgreSQL, independent application and authentication secrets, encrypted connector storage, HTTPS, backups, and ongoing upgrades.
- โณProvider behavior is not uniform: Railway computers currently require bearer authentication and private networking, while other remote providers impose their own API and usage requirements.
- โณThe Linux AppImage expects FUSE 2, and the desktop app does not detect Podman when searching for the `docker` CLI.
- โณThat changes the risk. A small team can operate the stack, but a casual user will face more deployment work than with a managed assistant.
Rakazo alternatives
Frequently asked questions
Rakazo is open source under the Apache License 2.0, with no paid license listed in the repository metadata. Model providers, voice services, and remote sandbox providers can still impose their own usage costs.
Rakazo can be used commercially under the Apache License 2.0 license, subject to its conditions. Self-hosters remain responsible for the terms attached to their chosen models, voice services, integrations, and remote computer providers.
Rakazo supports a bring-your-own-model approach through Pi by connecting credentials for a supported provider. The model agnostic design is a central feature: operators can select the model service instead of accepting one fixed assistant backend.
Rakazo includes an Electron desktop app and Expo applications for iPhone and Android. These clients connect to the same Rakazo API used by the web interface, so a self-hosted server can serve all three forms of access.
Rakazo can run on your own server by using the same published-image installer used for a local installation. Point the clients at the server address, place HTTPS in front of the application, and choose a remote sandbox provider when local Docker computers are unsuitable.
Rakazo is not presented as production-ready because the README labels it beta. It includes guidance for production deployment, HTTPS, backups, upgrades, monitoring, and audit logs, but operators still need to harden and maintain the installation themselves.
Who should try it โ and who should skip
Rakazo suits technical operators, privacy-conscious teams, and developers who want AI teammate software with persistent memory and actual computer access under their control. Skip it if you need a mature, fully managed assistant, a simple chat client, or an installation with almost no operational ownership. The sweet spot is a small team willing to maintain Docker, PostgreSQL, secrets, and provider access.
Related repositories
Still deciding about rakazo?
One click hands the question to an AI along with this page โ see what it says about rakazo.