# Rakazo: Self-Host the Open-Source Grok Bot Alternative > Source: > Published: 2026-10-25 07:08:55+00:00 xAI, OpenAI, and Meta each launched a persistent AI agent product in the past two months. Grok Bot, OpenAI Dots, and Meta Muse all follow the same pattern: an always-on AI with its own computer, memory, and tool access — locked to a single vendor, a specific model, and a monthly subscription ranging from $30 to $200. [Rakazo](https://github.com/elie222/rakazo), an open-source project by Inbox Zero creator Elie Steinbock, takes the same architecture and removes every one of those constraints. You pick the model, pick the sandbox, and host it yourself. It hit the top of GitHub Trending for TypeScript repositories on October 22 and currently sits at 5.6k stars. ## What Rakazo Is Each Rakazo bot is a persistent entity, not a chat session. It gets its own Linux desktop environment — a full Chromium browser, terminal, and filesystem — that stays running while you are away. Bots maintain their own memory and routines, can delegate tasks to peer bots or spawn subagents, and connect to external tools via MCP servers, Composio, Pipedream Connect, or OpenAPI definitions. The behavior matches what Grok Bot and OpenAI Dots offer, minus the vendor overhead. The stack is a TypeScript monorepo: React 19 and Hono on the front and back ends, PostgreSQL for persistence, Graphile Worker for background jobs, and Docker for the per-bot computer environments. There is also an Electron desktop app and an Expo mobile client for iOS and Android. All Apache 2.0. ## Model and Sandbox Choice Rakazo does not ship a model. During setup you connect whatever you use: OpenRouter (200+ models behind one key), OpenAI, Anthropic, Google, or a local server like Ollama, LM Studio, or llama.cpp. Each bot in your workspace can run a different model. A research bot on GPT-6, a coding bot on Claude Sonnet, a triage bot on a local Llama — none of that is locked. Sandbox options follow a similar tradeoff pattern: - **Docker (default)** — Free, local, fast. No external API key required. Weak inter-bot boundary: team bots share an OS user, workspace, browser profiles, and X11 access by default. Private bots get their own isolated container. - **E2B / Daytona** — Ephemeral cloud sandboxes with strong isolation. Traffic does flow through a third-party provider, which matters if you care about what those sandboxes can see. - **Railway** — Always-on servers at metered cost. Requires private networking configuration for the API and worker containers. ## Getting Started The quick install requires only Docker, curl, and OpenSSL — no repo clone, no Node install: ``` mkdir -p rakazo && cd rakazo && curl -fsSLO https://raw.githubusercontent.com/elie222/rakazo/main/infra/compose/install-images.sh && bash install-images.sh ``` Rakazo opens at `http://127.0.0.1:5173`. Create an account, connect a model provider during onboarding, and create your first bot. [OpenRouter](https://openrouter.ai/) is the path of least friction if you want access to multiple models without managing separate API accounts. ## What to Watch Out For Rakazo is beta software. The README is explicit about this, and breaking changes between releases are expected. A few specific issues are worth flagging before you hand a bot access to your tools: - **Prompt injection** : A bot browsing the web can be hijacked by malicious content on any page it reads. There is no built-in mitigation for this. - **Docker inter-bot isolation** : Team bots share an OS user and workspace by default. A compromised team bot can read another team bot’s files. Use private bots if isolation matters. - **Self-hosting is not the same as on-prem** : If your bots route prompts to OpenAI or Anthropic APIs, those requests still leave your infrastructure. Rakazo keeps your conversation metadata on your server, not your model traffic. - **Data persistence** : Running`docker compose down -v` deletes all PostgreSQL data. Export workspaces before tearing down. - **Podman not supported** : Graphical bot computers require Docker. Podman users are blocked by a known open issue. ## How It Compares The always-on agent category coalesced fast. Here is where Rakazo sits relative to the proprietary options: | Product | Provider | Price | Model | Self-hosted | |---|---|---|---|---| | Grok Bot | xAI/SpaceXAI | $30–$200/mo | Grok only | No | | OpenAI Dots | OpenAI | $100/mo (ChatGPT Pro) | GPT-6 Astra only | No | | Meta Muse | Meta | Free / $20–$100/mo | Muse Spark only | No | | Rakazo | Open source | Free (API costs) | Any | Yes | The persistent agent category is real and is developing fast. Grok Bot, Dots, and Muse represent one path — polished, managed, vendor-controlled. Rakazo represents the other — rough, configurable, and yours. For most individual developers and small teams, the beta warnings are worth accepting in exchange for not paying $100 per month to give an AI agent access to your tools. The [GitHub repository](https://github.com/elie222/rakazo) has a full setup guide, and the [ExplainX walkthrough](https://explainx.ai/blog/rakazo-open-source-grok-bot-alternative-github-trending-bring-your-own-model-sandbox-guide-2026) covers every sandbox and model option in detail.