Octop

// Overview

Octop is an open-source, self-hosted AI assistant platform built for households and small teams. A single process serves a web dashboard, a CLI, IM channels (Feishu, DingTalk, QQ, Discord, WeCom and more) and cron automation, all sharing one control-plane database under ~/.octop/ — SQLite with WAL by default, PostgreSQL optionally. The design goal is that every conversation, workspace and credential stays on your own machine while each user gets a personal team of specialised agents to switch between per task.

It is multi-user and multi-agent by default: one admin account with a shared household or team, a built-in expert library you pick a specialist from per scenario, and 16 MBTI persona templates (plus an interactive quiz) to give each agent a distinct character. Security is part of the model rather than an add-on — JWT multi-user isolation, per-tool approval, shell command guardrails and PII redaction. Backends are pluggable: local disk, Docker containers, PostgreSQL, or COS/S3, so agents operate inside an isolated boundary you choose.

Octop is composed from the Harness stack — harness-agent for the agent runtime, harness-gateway for the IM channel bridge, harness-memory for portable hierarchical recall that travels with a workspace, and harness-browser for CDP-based browser automation. On top of that sit a RAG knowledge base over your own documents, an OAuth and MCP connector gateway, a plugin system, a browser-based interactive terminal, headless Chromium sessions, remote desktop from the dashboard, and bidirectional ACP so you can delegate to (or be driven from) IDE and terminal agents.

// What it does

• One process serving a web dashboard, CLI, IM channels and cron, with all state under ~/.octop/
• Multi-user expert team with JWT isolation, tool approval, shell guardrails and PII redaction
• Connector ecosystem over OAuth and an MCP gateway, plus a plugin system for third-party extensions
• RAG knowledge base over your documents and portable hierarchical memory that migrates with the workspace
• Browser AI+ (headless Chromium for automation and screenshots), a web terminal and dashboard remote desktop on Linux, Windows and macOS
• Bidirectional ACP — expose Octop to IDE and terminal AI, or delegate tasks out to OpenCode and Claude Code behind permission gates

// Getting started

Octop runs on macOS, Linux and Windows. No preinstalled Python is required — the installer uses uv to provision Python 3.12 into an isolated virtualenv under ~/.octop/. Plan for a few GB of RAM for the process plus model and embedding caches.

curl -fsSL https://finnie-1258344699.cos.ap-guangzhou.myqcloud.com/octop/install.sh | bash

source ~/.zshrc   # or ~/.bashrc

# optional extras
curl -fsSL https://finnie-1258344699.cos.ap-guangzhou.myqcloud.com/octop/install.sh | bash -s -- --extras browser

pip install octop
# optional: pip install "octop[browser]"
# optional: pip install "octop[local-embedding]"

octop init

octop run

octop run --host 0.0.0.0 --port 8088

octop service start

docker compose -f docker/docker-compose.yml up -d

# or build and run manually
bash docker/docker_build.sh
docker run -d \
  -p 8088:8088 \
  -v octop-data:/data/.octop \
  -e HOME=/data \
  -e OCTOP_DEFAULT_PASSWORD="<strong-password-or-omit-for-random>" \
  octop:latest

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Running a household or small-team assistant entirely on your own hardware, with per-member accounts and agents
• Bridging Feishu, DingTalk, WeCom, Discord or QQ group chats into one agent pipeline instead of writing a bot per platform
• Grounding answers in a private document corpus while keeping every file and credential on the local machine
• Delegating coding work to OpenCode or Claude Code over ACP, or scheduling recurring jobs described in natural language

// How Octop compares

Octop alongside other open-source assistants & chatbots tools AI/TLDR tracks, ranked by GitHub stars.