Breaking

Nous Research Adds Search-Speed Features to Hermes Agent

July 1, 2026 at 03:15 EDT

AI Agents
Open Source

Hermes Agent, Nous Research's open-source self-improving AI agent, is drawing attention for optimizations that boost exploration efficiency without changing the underlying model. Alongside a new tool-search feature and faster web loading, users are reporting that simply restructuring the knowledge base's file layout can dramatically cut search times.

Feb 25, 2026 · Nous Research — Hermes Agent

Just tidying the folders made the AI agent 10× faster

No model swap, no extra compute — only adding INDEX.md navigation files and organizing the workspace. The agent stopped wandering and went straight to the target file.

10×

faster overall task processing

extra cost — same model, same compute

1 file

opened to reach the target (was ~7)

Time to reach the target file

Same unit (seconds) · column height drawn to scale

~120s

Before · ~2 min

~10s

After INDEX.md · ~10 sec

Only change made: place INDEX.md in each folder + tidy the structure.

60× faster

web loading, from the agent's own optimizations

49× cheaper

plus Tool Search loads only the tools needed

The closed learning loop — smarter the more you use it

Persistent memory

→

Autonomous skill creation

→

Self-improvement

Open-source (MIT) · 60+ built-in tools · 140,000+ GitHub stars within months

What works

Simple scaffolding sends the agent straight to targets
Dramatic speed gain at zero extra model cost
Memory + self-improvement compound over time

Watch out for

Heavy token use without caches / aggressive exploration
Initial setup and structuring effort required
Local models run more iterations than strong models

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.