Today for AI
HOT RADAR
agentHEAT 16.0°

AI CURATED HOT STORY · 10/29/2026

Infini-AI-Lab Open-Sources Hermit for Model-Agnostic Agent Learning

1 reports archived1 independent sourcesupdated 10/29/2026, 16:26:19
Synthesis & Latest Updates
1 Sources Cross-Validated

Infini-AI-Lab released Hermit, an Apache-2.0 framework enabling AI agents to learn continuously via a trainable 'Harness Adapter' without altering model weights. Using an RL loop and Git-based versioning for instructions and skills, it improved accuracy on CL-Bench and OfficeQA, though results rely on single runs and incur higher latency costs.

LATEST/Infini-AI-Lab released Hermit, an Apache-2.0 framework enabling AI agents to learn continuously via a trainable 'Harness Adapter' without altering model weights. Using an RL loop and Git-based versioning for instructions and skills, it improved accuracy on CL-Bench and OfficeQA, though results rely on single runs and incur higher latency costs.

Hourly heat curve

9 fully observed hours
Now
16.0
Peak
19.117:00
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first
  1. Marktechpost (@Marktechpost)T2·78 pts

    Infini-AI-Lab Open-Sources Hermit for Model-Agnostic Agent Learning

    • Core mechanism is the Harness Adapter: an external trainable layer of instructions, skills, memory, and tools that leaves model weights untouched.
    • Uses an RL-style loop: task execution -> judge scoring -> Evolver proposes new adapter version -> committed to Git.
    • Significant gains with trade-offs: GLM 5.3 Flash accuracy rose from 0.580 to 0.795 on CL-Bench, but at ~$0.16 cost and 7 minutes per question.
Infini-AI-Lab Open-Sources Hermit for Model-Agnostic Agent Learning | Today for AI