Infini-AI-Lab Releases Open-Source AI Agent Adaptation Framework Hermit

Techub News reports that Infini-AI-Lab has released Hermit, an open-source adaptation framework for AI agents. This framework allows developers to train an external "adapter" while freezing the agent's model and core program, enabling agents like Codex or Claude Code to adapt to new tasks without fine-tuning. The first version v0.1.0 was released on October 25, 2026.

Hermit uses a reinforcement learning-style training loop where adapters can include instructions, skills, memory, tools, and executable code. The current version offers five adapter evolution methods, including GEPA, SkillOpt, and MetaWorld Model. The team has reported performance improvements on benchmarks such as OfficeQA and CL-Bench.

Haizhong Zheng, a postdoctoral researcher at Carnegie Mellon University, announced this release on the X platform on October 27. (MarkTechPost)