DeepCode
About
DeepCode came out of HKU's Data Intelligence Lab as a multi-agent system whose orchestrator coordinates specialist agents for intent understanding, document parsing, code planning, reference mining, indexing, and generation, backed by CodeRAG and iterative verification. That pipeline reproduces machine-learning research papers as executable code, and the lab reports PaperBench results ahead of commercial agents on the commercial-agent subset. The project has since broadened into a general coding agent (v2.0) with a CLI/TUI and a Tauri-based desktop app sharing one runtime, full MCP client support with lazy loading and per-tool approvals, skills, plugins, subagents, and sandboxed sessions. Researchers use it for paper reproduction while developers adopt it as a general open-source harness; it is MIT-licensed, Python-based, and actively released.