Hugging Face· PlatformResearch paper

How Much Memory Does Your Agent Actually Need?

ALTK-Evolve is a method for agentic memory that distills guidelines from an agent's past trajectories and injects them at inference time without weight updates. The key finding is that the optimal amount of memory (guideline set size) depends on model capability: strong models benefit from the full set, weaker models perform best with a curated subset, and saturated models show no gain. For example, gpt-oss-120b gained +16.1pp task completion with curated retrieval at only +5% tokens.

From the source

The right dose differs by model tier: strong models with headroom want the full guideline set, weaker models do best with a compact core plus per-task retrieval, and saturated models show no measurable gain.

huggingface.co

Announced

18 Aug 2026 18:09 UTC

Detected