What is DeepCode?
DeepCode is an open-source agentic coding system built by the Data Intelligence Lab at the University of Hong Kong. It sits in the Autonomous Coding Agents category of the AI Coding tools and is open-source, written in Python and licensed MIT. To get started: install the CLI.
Overview
DeepCode is an open-source agentic coding system built by the Data Intelligence Lab at the University of Hong Kong. One agent runtime sits behind two interfaces: an interactive terminal CLI for people who live in a shell, and a Tauri desktop workbench that shows sessions, goals, tool activity, code changes and verification in a visual workspace. Both open the same local projects, session history, models, skills, permissions and automations.
Work is organised around durable sessions attached to a project. A session records tool calls and their results, not just the conversation, so resuming one restores what the agent actually ran; long conversations are compacted under context pressure while the recent tail survives verbatim. You can move the same history between the desktop app and the terminal without exporting or converting anything.
Models are yours to configure — you register a provider connection with an API key, refresh its catalogue and pick an exact model ID, so DeepCode is not tied to one vendor. The project also carries its research lineage as Paper2Code, the multi-agent architecture for turning research papers into working implementations that the repository grew out of, described in its arXiv paper. It is MIT-licensed and requires Python 3.12 or newer.
What it does
- One agent runtime behind two interfaces: an interactive CLI and a Tauri desktop workbench, sharing projects and history
- Durable sessions that record tool calls and their results, so a resumed agent can answer what it just ran
- Context compaction that keeps the recent tail verbatim and anchors on the token count the provider itself reports
- Bring-your-own models: provider connections you register, verify and switch per turn, with a thinking-effort control
- Reusable skills, durable goals and per-turn permission levels, settable from either interface
- Headless and automation modes (`deepcode exec`, goal runs) for scripted or scheduled work
- Paper2Code lineage: a multi-agent pipeline for reproducing research papers as code
Getting started
The CLI is the quickest way in. Install it with uv, register a model provider, then start the agent from inside the repository you want it to work on. The desktop app installs separately from GitHub Releases or from source.
Install the CLI
The explicit Python 3.12 selection is required — DeepCode must not fall back to an older interpreter.
uv tool install --python 3.12 deepcode-hku
deepcode initConnect a model provider
Register a connection once, load its model catalogue and verify inference. The --api-key flag opens a non-echoing prompt; credentials are stored separately and never written into session history.
deepcode provider set personal-openrouter --template openrouter --label "OpenRouter · Personal" --api-key
deepcode provider models personal-openrouter --refresh
deepcode provider test personal-openrouter --model <model-id>Start the agent in your repository
Run it from the project directory you want DeepCode to work in. Inside the session, /new starts a thread, /resume reopens history, /model and /effort change the next turn, and /permissions sets tool access.
cd <your-project>
deepcodeOr install the desktop workbench
Check GitHub Releases for a signed installer for your platform. Building from source needs the Tauri 2 prerequisites plus Git, Python 3.12+, uv, Node.js 22+ and stable Rust; then configure a provider under Settings → AI providers.
git clone https://github.com/HKUDS/DeepCode.git
cd DeepCode
uv venv --python 3.12
uv pip install --python .venv/bin/python -e .
.venv/bin/deepcode init
cd desktop && npm ci && npm run setup:sidecar && npm run build:sidecarCommands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Run a long, multi-step coding task where you need the session to survive a restart and still know which commands were executed
- Switch between a visual review workspace and a terminal on the same piece of work without losing history
- Point an agent at your own model provider and budget instead of a vendor-hosted coding assistant
- Reproduce a research paper as working code through the Paper2Code multi-agent pipeline
How DeepCode compares
DeepCode alongside other open-source autonomous coding agents tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| opencode | ★ 229k | OpenCode is an open source AI coding agent that runs in your terminal, with built-in build and plan agents and an optional desktop app. |
| Claude Code | ★ 157k | Anthropic's agentic coding tool for the terminal, IDE and GitHub: it understands your codebase, executes routine tasks, explains code and handles git workflows from natural-language commands. |
| OpenAI Codex | ★ 136k | OpenAI's cloud and CLI coding agent that writes features, fixes bugs, and proposes code changes across a repo, running tasks in parallel. |
| Orca | ★ 133k | Open-source desktop workspace that runs Claude Code, Codex, OpenCode and other CLI coding agents in parallel, each in its own git worktree, with a mobile companion app. |
| Pi | ★ 123k | Minimal terminal coding agent harness with four built-in tools, extended through TypeScript extensions, skills and prompt templates instead of forks. |
| Gemini CLI | ★ 109k | An open-source command-line AI agent from Google that connects your terminal to Gemini models for reading code, editing files, running shell commands, and searching the web. |
| autoresearch | ★ 100k | Karpathy's minimal harness that hands a coding agent a single-GPU LLM training script and lets it run fixed five-minute experiments, keeping or discarding each change on its own. |
| DeepCode | ★ 16.9k | An open agentic coding runtime with a terminal CLI and a desktop workbench |
