TencentCloud/Octop

A smarter, self-hosted AI assistant — multi-user, multi-agent.

What it solves

Octop provides a self-hosted, privacy-focused AI assistant platform that allows individuals, families, or small teams to run a multi-user, multi-agent system on their own hardware. It eliminates the need to rely on third-party cloud AI services for managing personal data, credentials, and conversation history.

How it works

Octop runs as a single process that integrates a web dashboard, a CLI, and various IM channel bridges. It uses a modular architecture based on the "Harness" stack:

  • harness-agent: Handles model routing, tools, and skills.
  • harness-gateway: Normalizes messages from different IM platforms (Discord, Feishu, etc.) into a single pipeline.
  • harness-memory: Provides hierarchical recall and full-text search for agent memory.
  • harness-browser: Enables web automation via headless Chromium.

All data, including a SQLite database and per-agent workspaces, is stored locally in ~/.octop/.

Who it’s for

  • Privacy-conscious users who want a fully self-hosted AI assistant.
  • Small teams or households needing a shared environment with isolated user accounts and specialized agents.
  • Developers who want to integrate AI agents into their IDEs or terminal via the Agent Client Protocol (ACP).

Highlights

  • Multi-Agent Ecosystem: Support for multiple agents per user, including MBTI-based personas and an expert library.
  • Omnichannel Access: Chat via Web UI, CLI, or IM platforms like Discord, DingTalk, and WeCom.
  • Remote Control: Includes a remote desktop feature for GUI apps and an interactive shell in the browser.
  • ACP Integration: Bidirectional support for the Agent Client Protocol, allowing Octop to act as a server for IDEs or delegate tasks to external coding agents like Claude Code.
  • Extensible Connectors: Supports OAuth and MCP gateways to extend agent capabilities.
  • Local-First Security: Built-in JWT isolation, PII redaction, and shell command guardrails.