# TokenWorth: Learning which memories are worth keeping under resource constraints

- Source: HuggingFace Daily Papers
- Published: 2026-10-19T00:00:00.000Z
- RoboHot: https://robohot.news/en/items/gsnajhjqg5ezilv39nd4r6kci
- Original: https://huggingface.co/papers/2610.26453

## Summary

The paper proposes TokenWorth, a lightweight online memory framework that learns query-relevant memory value from sparse execution feedback and applies it consistently to memory retrieval and bounded memory management. Across three LifelongAgentBench environments and in both local and API-based deployments, TokenWorth achieves the highest task success rate among the evaluated methods while using the fewest output tokens. The code is publicly available at https://github.com/AIGNLAI/TokenWorth.