OpenMausBot is a desktop chat app where every contact in the sidebar is a running agent. The original repository, milind-soni/OpenMausBot, sits at 4770 stars and is written in TypeScript under Apache-2.0. A Chinese translation of its documentation was published on 2026-10-14 at https://github.com/yangshun2005/OpenMausBot-cn.
That translation is what this post is about, plus what the project actually does and where it will bite you.
What the project is
The official one-liner is: "Open-source Grok Bot alternative with a virtual machine that bots can use."
The longer version, from the README: OpenMausBot is a local-first desktop chat application that turns a team of AI agents into something shaped like an instant messenger. Each bot in the sidebar is a real agent running Claude or Codex underneath, with its own personality, its own model, its own cloud computer, and its own connected apps. You talk to them like contacts. You watch them work. You approve the operations that matter.
The stated problem is that a single assistant in a single chat box does not fit agent workflows. OpenMausBot rebuilds the "AI as a messaging app" idea as an open-source, local-first project that runs on the agent CLIs you already have installed.
It describes itself as an open-source alternative to Grok Bot, and also to Meta Muse, OpenAI dots, and Cue by Manus. It is not affiliated with xAI, Meta, OpenAI, or Manus. It has also been called MausBot, and was previously named OpenGrokBot.
One thing the README states plainly, in a warning block: there is no token. Any coin using the OpenMausBot, Maus, or SupaMaus names was not created, endorsed, or associated with the project or its maintainers.
Why a Chinese translation matters here
The README is long. It covers install paths for three operating systems, Docker build targets, engine selection, host control permissions, and a pile of platform caveats. If English is not your first language, that is a lot of surface area to misread before you even get the app open.
The Chinese version splits the document into two parts, and the split is the useful idea:
- A 中文快速上手 section at the top, explicitly labeled as a supplement and not part of the original text. Its stated goal 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 ordering matters. You get the short path first, the reference material second. The translation also carries the original's warnings forward rather than dropping them, which is where a lot of translated docs go wrong — the caveats are usually the first thing to get trimmed.
The translator states that inaccuracies should be filed as an Issue or PR. My read is that the quick-start section is the part most likely to drift, since it is original content rather than a translation and will need updating whenever the upstream README changes. That is a guess on my part, not something the repo says.
Getting it running
The shortest path is a prebuilt installer. The README's download section lists:
- Mac, Apple silicon:
OpenMausBot.dmg - Mac, Intel:
OpenMausBot-intel.dmg - Windows:
OpenMausBot-setup.exe - Ubuntu 24.04 x64:
OpenMausBot-amd64.deb, or an AppImage beta
All releases are on the latest release page. The macOS builds are described as signed and notarized .dmg files.
If you want to build the harness server with Docker, the Dockerfile comments give these commands:
docker build -t openmausbot .
# pick your engines explicitly
docker build --build-arg ENGINES="@anthropic-ai/claude-code @openai/codex" -t openmausbot .
# build the cloud-home target
docker build --target cloud-home -t omb-cloud-home .
The Docker image is based on node:24-bookworm-slim.
For a source build, package.json specifies pnpm@10.33.0 and Node >=24. Be aware: the README material does not include a complete source build and run sequence. The Chinese quick-start says to consult the README body for that, which is honest but means you should not expect a copy-pasteable pnpm install && pnpm dev from the translation alone.
Requirements and the things that will stop you
The dependency situation is the part worth reading twice.
Bots run on CLIs you already have. OpenMausBot drives claude, codex, and grok installed on your own machine, using your existing logins and subscriptions. No new account. But this cuts both ways: if those CLIs are not installed and authenticated, the bots do not run. There is no bundled fallback described.
Platforms. macOS on both Apple silicon and Intel, Windows x64, and Ubuntu 24.04 x64. The badges note Electron support for macOS, Windows, and Ubuntu.
Host control is limited. Host control works only on macOS and Ubuntu Xorg, and only after you explicitly opt in. On Ubuntu Wayland it is disabled because of unresolved issue #345. If you are on Wayland and host control is the feature you came for, this is a blocker, not an inconvenience.
State lives in ~/.openmausbot. Transcripts, keys, and events go there. In Docker, HOME is the /data volume, which is what persists state across container restarts. Back that up if you care about your transcripts.
Hardware requirements are not specified. The README distinguishes Apple silicon from Intel installers and stops there. No RAM or disk figures. I would not read anything into that either way.
No mirror guidance. The material does not mention domestic npm mirrors or download accelerators. If you are behind a slow connection to npm or GitHub, that is on you to solve. The translation does not pretend otherwise.
Where the translation helps most
The quick-start section compresses the decision tree into a few lines: pick your OS, grab the installer, or build the harness with one of three Docker commands. It also front-loads the gotchas that would otherwise surface after twenty minutes of setup — the CLI dependency, the Wayland host-control gap, the state directory, the no-token warning.
For a Chinese-speaking developer evaluating whether this is worth an afternoon, that is the right shape. You can decide in two minutes whether the constraints disqualify you, then read the translated README for depth if they do not.
The honest caveats: the translation is new, published 2026-10-14, and the Chinese repository has no stars yet. It is a translation, not a fork with code changes, so the 4770 stars belong to the original project and say nothing about the translation's quality. If you find something wrong, the repo asks for an Issue or PR, and given how fresh it is, that is probably the fastest way to get it fixed.
Should you look at it
If you already run claude or codex locally and the idea of several persistent agents with separate personalities and separate cloud machines appeals to you more than one chat window, the project is aimed squarely at you. The local-first framing and the "use your existing subscription" model are the two design choices that make it worth a look rather than a dismissal.
If you are on Ubuntu Wayland, or you do not have those CLIs installed and do not want them, the constraints above are real and the Chinese quick-start says so up front.
Original project: https://github.com/milind-soni/OpenMausBot
Chinese documentation: https://github.com/yangshun2005/OpenMausBot-cn
Chinese documentation: https://github.com/yangshun2005/OpenMausBot-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)