Infini-AI-Lab Releases Hermit Harness Adapter Framework
The new open-source Python tool allows developers to train external adapters for AI models while keeping base components frozen.
Updated on Oct. 29, 2026 in Artificial Intelligence
Live Poll
Would you prefer to use external plugins to adapt your AI agents to new tasks?
Infini-AI-Lab has launched Hermit, an open-source framework designed to train external adapters for AI agents without modifying the original model or harness. Released under an Apache-2.0 license, the tool aims to improve adaptability and interoperability across closed and open-source systems.
Why it matters
Many existing agent harnesses are closed-source, and modifying them often disrupts critical component interactions. Hermit addresses this by enabling learned adapters to move between compatible models, preventing breakage while enhancing performance.
Hermit v0.1.0 supports five evolution methods including GEPA and SkillOpt, achieving a 45.3 score on ServeLearnBench hidden-policy tests. The framework is compatible with Codex, Claude Code, and Pi agents while maintaining adapter versions in a git tree.
The players
Infini-AI-Lab
This research entity specializes in developing open-source frameworks and tools for the artificial intelligence industry.
Haizhong Zheng
He is a developer and researcher who announced the release of the Hermit framework to the public on social media.
The details
The system operates by running tasks through an agent, where a judge scores outputs and an evolver proposes new adapter versions. Developers can use the git-based history to inspect versions or roll back changes if necessary.
Timeline
October 25, 2026: Infini-AI-Lab released the initial version of the Hermit framework.
October 27, 2026: Haizhong Zheng announced the availability of the framework on X.
The Tech Race
The framework utilizes the GLM 5.3 Flash model as a primary benchmark to demonstrate adapter effectiveness. By decoupling adapter training from model internals, this tool marks a move toward modular AI architectures that replace rigid, monolithic agent designs.
Developers can now implement modular upgrades to AI agents without the risk of breaking underlying system interactions. The git-based version control feature simplifies the workflow for tracking performance gains and managing experimental adapter versions.
The takeaway
Hermit offers a practical solution for developers struggling with the constraints of closed-source AI harnesses. Users can leverage the framework to test adapter performance across multiple agent platforms while ensuring system stability through standardized version control.
Further reading
For more developments in this field, visit our Artificial Intelligence section.
Source note: This article includes information reported by MarkTechPost.
Live Poll
Would you prefer to use external plugins to adapt your AI agents to new tasks?
