Guide · Open source

How to use DeepCode: install, run tasks and verify results

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

DeepCode is an open-source coding agent for real software projects. You give it a small change or a larger goal, and it reads code, edits files, runs commands and tests, and checks the results. This guide covers what you need to run it, how its three clients differ, and how to review its work. It assumes you are comfortable with a terminal and a software project.

DeepCodeHKUDS/DeepCode · ★ 16.8k · MIT Install options & downloads

Quick start

1Check Python. Confirm you have Python 3.12 or newer installed, since the README lists it as a requirement.
2Clone the repository. Run git clone https://github.com/HKUDS/DeepCode to download the source.
3Follow the README setup. Complete the remaining setup steps in the repository README.
4Pick a client. Use deepcode for the terminal, deepcode desktop for the desktop app, or deepcode web for the browser.
5Start a small task. Give the agent a scoped change in a real project you can test.
6Review the diffs. Inspect the diffs, test results and artifacts before accepting any change.

What you need before installing DeepCode

The README lists Python 3.12 or newer as the requirement. DeepCode is written in Python and is distributed as source from its GitHub repository, so you will clone it first.

It is aimed at developers and people comfortable with terminals. The README assumes familiarity with software projects, so it is not a no-code tool.

Python 3.12 or newer
Git, to clone the repository
A terminal

Get the DeepCode source code

Clone the repository from GitHub. The README's install section starts with this command. Follow the README after cloning for the remaining setup steps, since they are not reproduced here.

git clone https://github.com/HKUDS/DeepCode

Choose a client: TUI, desktop or web

DeepCode offers terminal (TUI), desktop and web clients. All three connect to one shared local service and share Projects, Session history, models, Skills, permissions, Goals and Automations, so you can switch between them without losing context.

The README names the launch commands for each client. Use the terminal client for keyboard-driven work, the desktop client for a visual workspace with Sessions, goals, tool activity and code changes, and the web client if you prefer a browser.

deepcode starts the terminal client
deepcode desktop starts the desktop client
deepcode web starts the web client

deepcode
deepcode desktop
deepcode web

Run your first task

Give DeepCode a task scoped to a real project. You can ask for a small change or a larger goal. The agent works through the project by reading code, editing files, running commands and tests, and checking results.

If the work is interrupted, or you restart or switch models, the README says DeepCode can continue the task. Start with a small, well-defined change so you can check each step.

Start with a small change you can verify
Watch the tool activity as it runs
Keep your tests or build command ready so the agent can check its work

Review diffs and task artifacts

DeepCode is built around verification rather than suggestions. The README describes checking results with tests, builds, diagnostics, Diffs and task Artifacts. Read the diffs before accepting changes, and use the artifacts to see what was done.

This is the main reason to pick DeepCode over a tool that only proposes code. Treat its output as work to inspect, not as a finished answer.

Read every diff before keeping changes
Check that the tests or builds the agent ran actually passed
Use task artifacts to trace what happened

Know the limits

DeepCode is open source under the MIT licence and written in Python. The README does not describe enterprise support, and it does not give guarantees for teams that need them. If you need documented support commitments, plan for that separately.

Researchers interested in agent harness and multi-agent orchestration can read the linked arXiv paper, which is the research origin of the project.

No documented enterprise support in the README
Requires Python 3.12 or newer
Non-programmers may find it hard to use without help

What people use it for

Making a small change to an existing project and having the agent run its tests
Working through a multi-step refactor with verification by builds or tests
Reviewing diffs and task artifacts for changes made by an agent
Experimenting with agent harness and multi-agent orchestration ideas from the linked paper

Frequently asked questions

Is DeepCode free?

DeepCode is released under the MIT licence, according to the README and repository metadata. Check the LICENSE file in the repository for the full terms.

What do I need to run DeepCode?

The README lists Python 3.12 or newer. You also need Git to clone the repository and a terminal to run the commands.

How do I install DeepCode?

Clone the repository with git clone https://github.com/HKUDS/DeepCode, then follow the remaining setup steps in the README.

Does DeepCode work on Windows, macOS or Linux?

The source does not list supported operating systems in the section used for this guide. Check the README for platform details before you install.

What are the alternatives to DeepCode?

Projects often compared with it include Aider, OpenHands, Cline and Goose. Pick based on whether you want a terminal tool, a desktop app or an editor extension.

Can DeepCode resume a task after a restart?

The README says it can continue after an interruption, restart or model switch. It also offers TUI, desktop and web clients that share one local service.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

DeepCode suits developers who want an open-source agent they can run on their own project and who will check diffs and test results. It is a poor fit for non-programmers, for anyone who cannot run Python 3.12 or newer, and for teams that need documented enterprise support.

Install options & downloads