Octop: TencentCloud's Self-Hosted Multi-User AI Assistant

TL;DR: TencentCloud/Octop is a Python, MIT-licensed, self-hosted AI assistant designed for multiple users and multiple agents with long-term memory. 3,490 stars on GitHub and a homepage at https://octop.cloud.

What it is

TencentCloud/Octop describes itself as “A smarter, self-hosted AI assistant — multi-user, multi-agent.” The repository is written in Python, has 3,490 stars, and is published under the MIT license. The homepage is https://octop.cloud and the GitHub topic list includes agent, agentic-ai, ai, ai-agent, ai-agents, local-first, and long-term-memory. The license field returned by the GitHub API is MIT, which means the project is free to self-host for commercial use without a separate agreement.

What it does well

The self-hosted AI assistant category is crowded, but Octop leans into three specific bets that show up in its topic list and its multi-user, multi-agent positioning.

• Multi-tenant by default. The assistant is designed to serve multiple users from a single deployment, so a team or a household can share one instance instead of running parallel single-user installs.
• Multi-agent capable. Multi-agent is in the description and in the topic tags, so the runtime is built around agents talking to agents rather than one chat box per user.
• Long-term memory. long-term-memory is a tagged topic, which signals a persistent memory store rather than a context-window-only model. That matters once a single user has weeks of conversation history to recall from.
• Local-first operation. local-first is a topic tag, so the default deployment shape assumes your machine or your VPS, not Tencent Cloud.

How to try it

The homepage is https://octop.cloud; the quickest path is the docker compose the repo ships with.

# Clone and start the stack
git clone https://github.com/TencentCloud/Octop.git
cd Octop
docker compose up -d

# Open the local UI
open http://localhost:8080

From there the assistant is wired to whatever model provider you configure in the env file; the repo’s README documents the supported providers.

When it fits, and when it doesn’t

Octop fits teams or households that want one self-hosted assistant serving several people, with persistent memory that survives across sessions and the option to run agents on top of that memory. It is less attractive if you want a single-user local box with no auth — anything lighter will do — or if you are happy with a hosted SaaS assistant and do not need to keep the conversation store on your own hardware. The 3,490 star count is modest but the topical alignment (multi-user, long-term-memory, local-first) is unusually clean, and the homepage being on a .cloud domain while the code is local-first suggests the project is positioned as self-hosted with an optional managed control plane rather than a pure SaaS play.

Source and Accuracy Notes

• Project page: https://github.com/TencentCloud/Octop
• Source repository: TencentCloud/Octop (verified via GitHub API html_url)
• License: MIT
• Stars at time of writing: 3,490
• Language: Python
• Topics: agent, agentic-ai, ai, ai-agent, ai-agents, local-first, long-term-memory
• Homepage: https://octop.cloud