DeepCode
About this tool
HKU Data Intelligence Lab's open-source multi-agent coding system, originally Paper2Code. It reads papers, documents or repositories, plans an implementation and produces runnable reproduction code. The README reports 75.9% on PaperBench's human-expert subset. It now supports general goal-driven coding with analysis, implementation, validation and repair loops.
- Feature 1: Paper2Code generates reproduction code from papers, URLs or repositories.
- Feature 2: Goal-driven implementation, testing and repair loops.
- Feature 3: CLI, web (deepcode web) and desktop interfaces.
- Feature 4: stdio, SSE and Streamable HTTP MCP servers, with agent skill extensions.
- Feature 5: OpenRouter, OpenAI, Anthropic and DeepSeek, or local Ollama and vLLM models.
Resource information
- Maintainer
- HKUDS, the Data Intelligence Lab at the University of Hong Kong (HKU), not HKUST; official university-laboratory project.
- Type
- Standalone app
- Category
- Research agents & deep research
- Runs on
- Agent application
- Who it's for
- ML graduate students and engineers reproducing algorithms, building baselines and practicing paper implementations for submissions, meetings or courses.
- License at time of listing
- MIT
- Source
- GitHub
