The upstream repo elie222/rakazo has 3869 stars and is written in TypeScript. Its own one-line description is "Open-source Grok Bot alternative. Choose your own model and sandbox." What it does not have is a Chinese README. Someone fixed that: yangshun2005/rakazo-cn is a translation of the original documentation, published 2026-10-14.
This article is about what rakazo actually is, how the translated docs are structured, and where the Chinese version helps versus where it just mirrors the English.
What rakazo is
Rakazo is an open-source AI agent positioned as a persistent AI teammate — a self-hosted assistant that runs on your own machine. It ships clients for web, Electron desktop, iPhone, and Android. You can bring your own model and computer provider, or run the entire stack locally.
The comparison set the upstream README names is specific: Grok Bot, Meta Muse, OpenAI Dots, Instinct, OpenClaw, Hermes Agent, and Hark Pro. Rakazo is pitched as the open-source, self-hostable alternative to that group. Some of those are hosted assistants (Muse, Dots, Instinct, Hark Pro); some are open-source agents you run yourself (OpenClaw, Hermes Agent). The distinction the README draws is that rakazo, once running, is just a chat — you set up a bot and manage it through that chat.
The project is in beta. The README says so explicitly, and that matters for anyone deciding whether to put it in front of real work.
License is Apache-2.0.
Why a translation is not just cosmetic
Self-hosted software has a documentation problem that hosted software does not. With a hosted product, a bad README costs you a signup. With self-hosted software, a bad README costs you an evening of debugging environment variables, and the failure modes are silent — a container that starts but cannot reach a model, a .env where two values must match and one of them doesn't.
The rakazo README is long. It covers install paths, provider configuration, secrets, and a self-host guide. For a Chinese-speaking developer evaluating the project, that is a lot of English prose to parse before you know whether the thing even fits your use case. The translated repo splits the problem in two:
- A 中文快速上手 section at the top, marked as supplementary and not part of the original text. Its stated purpose is to let you get the project running without reading the whole README.
- The translated README body below it, which aims to stay faithful to the original.
That separation is the right call. Mixing editorial guidance into a translation makes it impossible to tell what upstream actually said. Keeping them apart means you can trust the lower half as a translation and treat the upper half as a getting-started guide.
The shortest install path
The translated quick-start gives the published-image route, which needs no clone and no Node install. Prerequisites are Docker Engine, the Compose plugin, curl, and OpenSSL.
mkdir -p rakazo && cd rakazo &&
curl -fsSLO https://raw.githubusercontent.com/elie222/rakazo/main/infra/compose/install-images.sh &&
bash install-images.sh
The script downloads the Compose files, generates a .env with random secrets, and starts rakazo. Rerunning it preserves an existing .env, which is the behavior you want — you don't lose your keys on a second run.
Then open http://127.0.0.1:5173, create an account, and connect a model.
For a server deployment, the flow is three steps instead of one: run bash install-images.sh --prepare-only, edit .env to set SANDBOX_PROVIDER and RAKAZO_HOST, then run bash install-images.sh again.
The default image tag is edge, built from main. Published images cover linux/amd64 and linux/arm64.
Requirements, and the Node version trap
If you go the published-image route, the dependency list is short: Docker Engine, Compose plugin, curl, OpenSSL. The README does not specify an OS or hardware requirement.
If you build from source, the Node constraint is worth reading twice. Supported: Node.js 22.22.2 or higher within the 22.x line, Node.js 24.x, or Node.js 26 and above. Not supported: Node.js 23.x and 25.x. Package manager is pnpm 9, plus Docker.
That odd-numbered-release exclusion is the kind of detail that costs an hour if you skim. It is also exactly the kind of detail that a translated quick-start section can surface prominently, because the translator is reading the doc as a user, not as the author.
Secrets you will have to set by hand
The published-image path generates secrets for you. Source development does not. The README lists what you must configure:
-
POSTGRES_PASSWORD, which also has to be reflected inDATABASE_URL BETTER_AUTH_SECRETENCRYPTION_KEYSCREEN_PROXY_SECRET-
SANDBOX_SUPERVISOR_TOKEN, needed for the Docker sandbox
The POSTGRES_PASSWORD / DATABASE_URL pairing is the classic failure: set one, forget the other, get an auth error that doesn't name the real cause.
Model credentials and remote providers
Rakazo needs a model. You can set OPENROUTER_API_KEY, or connect a supported model provider during onboarding.
Voice mode needs its own keys — the README names ElevenLabs, OpenAI, Cartesia, and Fish Audio as options.
The local Docker computer is enabled by default. If you want a remote provider instead, the options are e2b, daytona, createos, or box, each requiring its own API key.
The restricted-network section
This is where the translation earns its keep for a specific audience. The README documents overrides for constrained networks:
-
RAKAZO_DOWNLOAD_BASEto change the base URL the install script downloads from -
--localorRAKAZO_DOWNLOAD_SKIP_EXISTINGto skip Compose files you already have - Overriding the bootstrap script URL
The details live in the self-host documentation under "Restricted networks / mirror downloads." If you have ever tried to pull a GitHub raw file from a network that rate-limits or blocks it, you know why this section exists. Having it in Chinese, with the environment variable names left in English so you can copy-paste them, is the correct handling.
What the translation does not do
It does not change the software. Same repo, same images, same env var names. The Chinese version is documentation only.
It also does not claim adoption. The translated repo is new — published 2026-10-14 — and has no stars to speak of. The 3869 figure belongs to the upstream project, and the translated README lists it that way, under "原项目 Stars." If you see the number quoted elsewhere as if it described the translation, that's a misread.
The translation also flags its own limits. The header states that the quick-start is supplementary and not original content, and invites issues or PRs for inaccuracies. That's a reasonable posture for a translation of a beta project's docs — upstream will change, and the translation will lag.
Should you use it
If you read English comfortably, go to the upstream README. It is the source of truth and it will be current.
If you are evaluating rakazo and English documentation is the friction point, the Chinese version gives you two things the original does not: a condensed path to a running instance, and a translated reference you can search in your own language when something breaks. The env var names stay in English, which is right — those are the strings you actually type.
The project is beta. The translated docs say so. Treat both accordingly.
Repo: https://github.com/yangshun2005/rakazo-cn
Upstream: https://github.com/elie222/rakazo
Chinese documentation: https://github.com/yangshun2005/rakazo-cn
I'm an independent enterprise AI delivery consultant — private LLM deployment, RAG, agent workflows, AIOps. If your team is stuck somewhere in that stack, my inbox is open.
Top comments (0)