HugAgentOS
A self-hosted AgentOS for one workspace: plan-mode chat, private knowledge base, MCP tools, skills, memory and a sandbox
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- Runs code in a sandbox and carries out tasks the agent decides on
- Hosts a network service with a database and file storage
- Connects external models and MCP tools and accesses the network
Install
Manual install
curl -fsSL https://raw.githubusercontent.com/ZJU-REAL/HugAgentOS/main/install.sh | bashOne-command install on Linux, macOS or WSL2. Requires Python 3.11+, Node.js 20+, Git and curl; no Docker needed. The service opens at http://127.0.0.1:3001.
This is third-party code. Review the repository files before installing.
What it does
HugAgentOS is a self-hosted platform that brings agentic chat, a private knowledge base with RAG, sub-agents, MCP tools, skills, a code sandbox, long-term memory and a data canvas into one workspace. The core is built on AgentScope 2.0, a ReAct loop and the Model Context Protocol, so the agent can search, analyze, produce files and call external capabilities. It includes automation and batch runs: tasks are created in natural language and one process can run across rows of a table or a file list. A model connects through a single configuration; cloud or local OpenAI-compatible providers work, and data and storage stay on your own infrastructure. The Community Edition is open source, while some features such as team collaboration and persistent sandboxes are reserved for the Enterprise Edition.
Who it is for. For developers and teams who want a self-hosted AI agent workspace on their own server rather than a cloud service.
Good fit when
- You need to keep the agent's data and history on your own infrastructure
- You want chat, a knowledge base, skills and a sandbox in one place
- You want to connect a local or OpenAI-compatible model without vendor lock-in
Not a fit when
- You need a plugin or skill inside an agent you already use, not a separate platform
- You have no resources to self-host and maintain a server
Example request
Run this report across every row of the table and build a summary in the knowledge baseLimitations
This is a self-hosted platform: you need your own server and an external or local OpenAI-compatible model. Install requires Python 3.11+, Node.js 20+, Git and curl; the Docker Compose path adds PostgreSQL and Redis. The initial login and password are both admin and must be changed on first sign-in, and the Community Edition has no self-registration. By default the service listens on 127.0.0.1 only, and remote access must be set up manually with a password, firewall and HTTPS. The license is Apache 2.0 with supplementary terms: you may not run it as a competing multi-tenant SaaS, and the UI's Powered-by attribution must stay visible. Some features are Enterprise-only.
How to disable. For the one-command install, stop the hugagent process and remove the ~/.hugagent directory. For the Docker install, run docker compose down in the project folder and delete the volumes and the repository directory.
Security check
- Runs code in a sandbox and carries out tasks the agent decides on
- Hosts a network service with a database and file storage
- Connects external models and MCP tools and accesses the network
README in short
The README describes HugAgentOS as an enterprise-grade AgentOS that raises domain ontology into a control plane for the agent's reasoning and actions. The open-source Community Edition combines agentic chat, private knowledge-base RAG, sub-agents, MCP tools, skills, a sandbox, long-term memory, automation and a data canvas on one server. Two install paths are given: one command on SQLite and Docker Compose with PostgreSQL and Redis. It also details the technology stack, a layered architecture, the Community versus Enterprise boundary and a roadmap. The Community repository is generated from the upstream one at each release, under Apache 2.0 with supplementary terms.
FAQ
Is Docker required?
No. The one-command install runs the service on SQLite and a local sandbox without Docker. Docker Compose is for when you need PostgreSQL, Redis, an isolated sandbox and persistent volumes.
Which model do I need?
Any cloud or local OpenAI-compatible model, connected in settings through a single model-service configuration.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent