# HugAgentOS

A self-hosted, open-source AgentOS that combines agentic chat, private knowledge-base retrieval, sub-agents, MCP tools, sandboxed execution and long-term memory in one workspace, using domain ontology to guide agent reasoning.

HugAgentOS treats domain ontology as a control plane for how its agents reason, decide and act. Its open-source Community Edition combines agentic chat, private knowledge-base retrieval-augmented generation, sub-agents, MCP tools, agent skills, sandboxed code execution, long-term memory, automation and a data canvas within one self-hosted workspace, with skills, memory and orchestration each able to evolve and take effect only after a user approves the change.

The Community Edition installs on Linux, macOS or WSL2 with a one-command installer needing Python 3.11 or newer, Node.js 20 or newer, Git and curl, with no Docker, PostgreSQL or Redis required, using SQLite and a local subprocess sandbox that the project describes as suited to personal use and development; a Docker Compose path is offered for a longer-running, isolated deployment instead. Either way it needs an OpenAI-compatible or local model to talk to, and the initial admin account and password must be changed on first sign-in, since the Community Edition has no self-registration.

By default the installer listens only on localhost, and the project warns that exposing it more widely needs an explicit host and port flag plus a strong password, a firewall and HTTPS. It is written in Python; this Community repository is regenerated from an upstream repository on each release.

Website: https://github.com/ZJU-REAL/HugAgentOS