HKUDS

Generates command-line interfaces for GUI software so AI agents can drive it, with a hub of community CLIs.

Multimodal RAG framework that ingests text, images, tables, and equations from documents into one query pipeline.

Lightweight open-source agent that works across tools, chats, and workflows, integrating with Claude and Codex.

Retrieval-augmented generation system combining vector search with graph-based knowledge representation.

Agentic coding system from HKUDS that turns research papers and text descriptions into working frontends and backends.

Framework that plugs into existing AI agents so skills improve from real usage, cut token costs, and share learning.

Zero-code LLM agent framework from HKUDS where agents are defined in natural language, with no Python required.

Speeds up code understanding with LLM-powered repository navigation, symbol mapping, and concept extraction.

Agentic framework covering video understanding, editing, and remaking in a single open-source toolkit.

RAG system for video that lets you chat with hours of footage and get cited timestamps and clips back.

Autonomous research agent from HKUDS that proposes questions, runs experiments, and writes up findings end to end.

ACL 2026 code and training recipes for testing whether small language models can teach large ones to reason.

Universal tool-use layer that orchestrates arbitrary tools and MCP servers for computer-use agents.

Agent-driven personal trading assistant built in Python and deployable with Docker.

Research assistant agent that suggests hypotheses and novel ideas for a given scientific domain.

ICML 2025 method that speeds LLM inference by compressing each text segment into a single separator token.

Fully automated, agent-native trading system powered by large language models, written in TypeScript.

Self-supervised learning framework for recommender systems spanning contrastive, generative, and graph methods.

Video-generation system staffed by agentic director, screenwriter, and producer roles.

Recommendation framework from a WSDM 2024 paper that augments graphs with LLMs and multi-modal side signals.

Personal AI agent that learns from your screen and recall history to deliver context-aware responses.

Python framework for LLM agents covering tool use, computer use, and multi-agent orchestration.