GitHub radar
Octop: a self-hosted AI assistant for your home or team
Octop is an open source AI assistant you install and run yourself on your own machine, serving several users and several "expert" agents at once.
It's a self-hosted AI assistant that runs on your own computer instead of someone else's cloud. You can keep several "expert" agents for different tasks and switch between them.
When it helps
Useful if you want one assistant across all your chats — Telegram, Discord, a web dashboard — instead of a separate bot per app, especially for a family or small team. Skip it if one simple chatbot already covers your needs.
Pros
- Runs on your own machine — data stays off someone else's cloud
- One assistant across every chat app: Telegram, Discord, web dashboard
- Several experts for different tasks sharing memory
- Supports multiple users — works well for a family or team
Cons
- No model of its own — you must connect one
- Only as good as whatever model you connect
- AgentTeams and some features are still in beta
How to set it up — step by step
- 1Open your AI agent (e.g., Claude Code) inside your project folder.
- 2Send the agent the text from the block below — it installs Octop from GitHub and sets everything up.
- 3When the agent asks for a model key, give it a DashScope key (Alibaba's cloud service hosting Qwen models) or set up Ollama (free local model runner) instead.
- 4Ask the agent to set up one expert for your task.
- 5Check the result: open the web dashboard (usually http://127.0.0.1:8088) and message the expert on Telegram.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Deploy Octop from https://github.com/TencentCloud/Octop — follow the README (one command installs everything via uv), connect a model with an OpenAI-compatible key or through Ollama, set up one expert for my task, and let me message it on Telegram. Ask me for a model API key if you need one.
Octop itself is light — a modern CPU and a few GB of RAM for the process and caches, no GPU needed. If you want the model fully local through Ollama, that model needs its own resources depending on its size; with a cloud key, a regular computer or a cheap server is enough.
Open on GitHub▌ More finds