Open-source long-horizon super agent harness that orchestrates sub-agents, memory, and sandboxes to research, code, and create across tasks lasting minutes to hours.
HKUDS/DeepCode
An open-source agentic coding tool with TUI, Desktop, and Web clients that share one local service, for working on real repositories with multi-agent orchestration.
Open-source alternative to:Claude Code
Cursor
- Stars
- 16,736
- Forks
- 2,174
- Open issues
- 35
- Last commit
- 6 days ago
What is it?
DeepCode is an open agentic coding framework that lets an AI agent explore a codebase, edit files, run commands and tests, review changes, and carry a task through to a verified result. Its design centers on a goal-driven loop: the agent keeps working through understanding, implementation, verification, and repair rather than stopping after a single answer, and the user can add requirements, switch models, stop, or resume without discarding completed work.
The terminal UI, desktop app, and web workbench all connect to one shared local background service, so projects, session history, models, skills, permissions, and automations are consistent across interfaces. Sessions persist to disk, and tool activity, plans, code changes, and test results stay visible for review. Users can bring their own model providers, including local model servers, and extend the agent with reusable skills and MCP servers.
The project also includes Paper2Code, a workflow for turning research papers into runnable code, and supports headless execution for scripted or scheduled runs.
Key features
- 01Terminal, desktop, and web clients share one local background service.
- 02Goal-driven loop that continues through understanding, implementation, verification, and repair.
- 03Sessions persist to disk and can be resumed, with tool activity and test results visible.
- 04Supports user-configured model providers and per-session reasoning settings.
- 05Extensible through reusable skills, MCP servers, and plugins.
- 06Headless execution and recurring automations for repository maintenance tasks.
Use cases
- Delegating bug fixes, test repair, or refactoring tasks to an agent working inside a local repository.
- Turning a research paper into a runnable code implementation with Paper2Code.
- Scheduling recurring checks such as regression tests or documentation upkeep.
- Experimenting with multi-agent coding workflows using your own model providers.
Quick start
- 1
Install the runtime (follow the Install the runtime section of the Quick start)
- 2
Configure a model (follow the Configure a model section)
- 3
Start the terminal UI
deepcode - 4
Or start the Web client
deepcode web
Things to know
- Requires Python 3.12 or newer.
Topics
FAQ
- What is DeepCode?
- An open-source agentic coding tool with TUI, Desktop, and Web clients that share one local service, for working on real repositories with multi-agent orchestration. DeepCode is an open agentic coding framework that lets an AI agent explore a codebase, edit files, run commands and tests, review changes, and carry a task through to a verified result. Its design centers on a goal-driven loop: the agent keeps working through understanding, implementation, verification, and repair rather than stopping after a single answer, and the user can add requirements, switch models, stop, or resume without discarding completed work.
- Is DeepCode free to use? What license does it have?
- DeepCode is open source under the MIT license. Check the license terms before using it in commercial products.
- Is DeepCode still maintained?
- Actively maintained. The last commit to DeepCode was on Oct 14, 2026.
- What can I use DeepCode for?
- Delegating bug fixes, test repair, or refactoring tasks to an agent working inside a local repository. Turning a research paper into a runnable code implementation with Paper2Code. Scheduling recurring checks such as regression tests or documentation upkeep. Experimenting with multi-agent coding workflows using your own model providers.
- How do I install DeepCode?
- Install the runtime (follow the Install the runtime section of the Quick start) Configure a model (follow the Configure a model section) Start the terminal UI: deepcode Or start the Web client: deepcode web
- Where can I get DeepCode?
- The source code and installation instructions are on GitHub: https://github.com/HKUDS/DeepCode
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
An open-source general AI agent framework that runs LLM-driven agents from a terminal, with tool use, MCP support and browser automation.
Self-hosted personal AI agent framework in Python with a browser WebUI, terminal client, chat app channels, tools, memory and MCP support.
An opinionated, batteries-included agent harness for Python that runs long multi-step tasks out of the box on LangGraph.
A minimal Python library for building AI agents that write and run their actions as Python code.
Python framework for building autonomous AI agents that research, plan, code and run tasks, with memory, RAG and 100+ LLM providers.