Octop is a self-hosted, multi-user AI assistant from Tencent Cloud that runs as a single process on hardware you own. It is built for a household or a small team rather than one person: an admin account creates several human users, and each of them gets their own agents, personas and credentials inside the same installation. The people it suits are those who already pay for a cloud assistant subscription, want the same kind of help, and are willing to keep a small machine running at home to get it for free.
What it does
Octop gives a group of people a shared assistant with individual accounts behind it. One admin sets up users; each user configures their own agents and personas, so a family of four does not share a single chat history or a single assistant personality.
You can reach it three ways:
- a web dashboard, where you watch agents work;
- a command-line interface;
- the chat apps people already keep open — Feishu, DingTalk, QQ, WeChat, WeCom, Telegram and Discord.
That last route is the distinctive part. Octop has no app of its own to install on a phone; it bridges into seven-plus messaging platforms at once from one deployment, so talking to your assistant means sending a message in a client you already use. It also runs scheduled, cron-style automation, which means agents can do work on a timer rather than only when someone asks.
How it works
The architecture is deliberately small. Everything — the dashboard, the CLI, every chat-platform bridge and the scheduled automation — runs in one Python process backed by one database, either SQLite or Postgres. There is no Redis, no message broker and no separate job queue to stand up alongside it.
That design has an obvious upside for a home install: one thing to start, one thing to back up, one thing to debug. It is meant to sit on a laptop, a home server or a NAS rather than in a cluster. The trade-off is just as obvious. A single process is a single point of failure, and it scales by giving the box more resources rather than by adding more boxes. For a few people in one house that is the right call; for an organisation it is a ceiling.
Because the process runs on your machine, conversations and configuration stay in your own database instead of a vendor's account. The project is MIT licensed, so you can read it, change it and run it without asking anyone.
Getting started
Installation, by the channel's own account, is a single command, after which you open the built-in dashboard and start adding users and agents. Pick the chat platforms you care about and connect the bridges you need; you do not have to enable all of them.
Two numbers are worth knowing before you commit. The repository is at roughly 6.3k stars and 788 forks, with hundreds of open issues and pull requests and daily commits, so it is actively worked on and actively reported against. The version is 1.0.2b5 — a beta label. Treat it as software that is moving quickly rather than something settled.
When to use it / when not
It fits well when several people share one AI setup and each wants their own agents; when you would rather reach an assistant from WeChat, Telegram or Discord than from a separate app; when the data ought to stay on your own disk; and when you want recurring tasks to run on a schedule without renting anything.
It fits badly when you want a finished consumer product — this is a self-hosted project at a beta version, with the rough edges that implies. It fits badly if nobody in your group uses any of the bridged platforms, since the chat integration is most of the appeal. And it fits badly if you have no appetite for running and updating a server process yourself; a subscription exists precisely so that you do not have to.
Alternatives
The honest comparison is ChatGPT or any other paid cloud assistant: one account, one assistant personality, data on someone else's server, a monthly bill. Octop trades the polish of those services for multi-user accounts, local storage and no fee.
Within the self-hosted world, Octop is closest in spirit to the various "run your own ChatGPT" projects. What sets it apart there is the combination of real multi-user separation and bridges into many chat platforms from a single deployment, rather than one web UI for one person.
Take Octop seriously if you are the person in a household or small team who already maintains a NAS or home server and is tired of paying per seat for an assistant that everyone shares badly. The single-process design means the operational cost of trying it is low, and the MIT license means nothing locks you in. Anyone who wants a product that simply works, with support behind it, should keep paying the subscription — this is a beta you run yourself, and that is the deal on offer.