AI NewsOpen sourceAnnouncement
Rakazo, a self-hosted platform for AI teammates, reached 1,822 stars
Rakazo is a self-hosted platform for persistent AI teammates with bring-your-own-model credentials and isolated computer sandboxes, and has picked up 1,822 GitHub stars in the three weeks since it was created.
Image: GitHub
Why it mattersTeams that want the features of a hosted agent product without giving a vendor their code, credentials or model choice now have a working self-hosted option with real adoption behind it.
Rakazo, an open-source platform for running persistent AI teammates, has reached 1,822 stars and 308 forks on GitHub in the 21 days since its first commit on 13 August. The repository is Apache 2.0, written mainly in TypeScript, and the project's own description calls it an open-source alternative to Grok Bot with a bring-your-own-model and bring-your-own-sandbox design.
What the project includes
The README describes Rakazo as a full stack: a web app, an Electron desktop client and an Expo mobile app, all backed by a Hono and oRPC server, PostgreSQL through Prisma, Better Auth and Graphile Worker. Bots have their own conversations, memory, routines and history, and can delegate work to peer bots or to short-lived subagents. A voice mode is included, with the user supplying an ElevenLabs, OpenAI or Cartesia key.
The computer layer is where the difference from hosted products is clearest. A bot can be given a shared Team Computer or an isolated Private one, and the supported providers listed in the README are Docker, E2B, Daytona, Box and a trusted local computer. Each provider gives the bot browser, terminal, file and graphical desktop access.
How integrations connect
Tool sources are Composio and Pipedream Connect for managed catalogs, plus user-installed Treg endpoints, remote MCP servers and OpenAPI documents that a user can add from the Integrations screen without either managed catalog being enabled. Connector credentials are encrypted server-side and the README states they are never returned by the API. Model credentials are provided through a component the project calls Pi, and can also be set from an OpenRouter key at onboarding.
What it costs to adopt
The published-image install path is Docker Engine plus the Compose plugin: one shell script pulls the images, writes an .env with random secrets and starts the stack. Local Docker computers are on by default, and the E2B, Daytona and Box providers are opt-in through their API keys. Treg is usage-metered and the README notes that self-hosters supply their own token, while operators embedding Treg in a hosted product are pointed at Treg's separate integration terms.
The interesting part here is who the project is for. A team that wants the ease of use of a hosted agent product, persistent bots with memory and their own sandboxes, but does not want to hand their code and credentials to a vendor, has had few honest options. Rakazo is one, and the growing star count shows people are interested. Twenty-one days is early, and the repository is still tagged as beta on its own website, so treat this as a project to try in a sandbox rather than an approved one for production data.
The design choices are the part worth reading before adoption: model provider, sandbox provider and tool source are each swappable, so a team is tied only to the schema and the frontend, and stays free to change the model or the sandbox operator. That is the difference between running your own teammate and renting one.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.