A Unified Framework for Training, Evaluating, and Deploying GUI Agents

Full-stack framework for GUI agents: online RL training in parallel emulators, standardized evaluation across 6 benchmarks, and one-command deployment to real Android, HarmonyOS, and iOS devices.

ClawGUI system architecture
ClawGUI-Agent controls a real phone via natural language
ClawGUI-RL trains a GUI agent with online reinforcement learning
Contributors
View all →
A Unified Framework for High-Performance and Extensible LLM Steering

High-throughput LLM activation steering built on vLLM. Apply semantic intervention vectors at inference time without modifying weights — 10.8–22.3× faster than prior frameworks, with pre-computed vectors for 8 domains and an interactive web demo.

EasySteer architecture
EasySteer demo — interactive vector testing and multi-turn chat
Contributors
View all →