DeepCode is an open agentic coding system from the University of Hong Kong’s data science group (HKUDS), built around multi-agent orchestration, agent harness design and loop engineering. It made its name turning research papers and ideas into working code repositories, and it has grown into a full coding workspace: a terminal UI, a desktop app and a web client all run over one shared local service, sharing projects, session history, models, skills, permissions, goals and automations. DeepCode has earned more than 16,000 stars on GitHub and is under active development in 2026.
Where many coding agents are a single model in a loop, DeepCode treats coding as an orchestrated process: specialized agents plan, implement and verify, coordinated by a harness that manages context, tools and permissions across long tasks. You can drive it from whichever interface suits the moment, the terminal for quick work, the desktop app for a visual workspace showing sessions, goals, tool activity, code changes and verification, or the web client, all against the same local service. Skills and automations let you package repeatable workflows, and a permissions system keeps agent actions under your control.
DeepCode is free and open source under the MIT license. You bring your own model access, so your only costs are any LLM API keys you choose to use.
DeepCode is for developers and researchers who want an open, multi-agent coding system they can run locally, especially for long, structured tasks like reproducing papers or building features across many files.