TokenWorth scores memory worth for lifelong agents under tight budgets
Original title: Remembering Has a Price: Learning Memory Worth under Resource Constraints
Across three LifelongAgentBench settings it records the highest success while emitting the fewest output tokens of the methods tested.
The paper argues that under tight generation, interaction, and memory budgets, several lifelong-memory methods offer little gain over simple retrieval or no-memory agents. TokenWorth learns query-dependent memory worth from sparse execution feedback and applies it to retrieval and bounded memory management. Evaluations cover three LifelongAgentBench environments and both local and API-based deployments. Code is public.
- A relevant but ineffective memory can consume scarce context and hurt execution.
- Worth estimates come from sparse execution feedback and stay query-dependent.
- The same signal governs retrieval and bounded memory retention.
- Tests include local and API-based deployments.
From the original article
1 min read at huggingface.co
Lifelong agents continually accumulate and reuse experience to improve after deployment, yet in practice they often operate under tight generation, interaction, and memory budgets. Under such stringent joint constraints, we find that existing lifelong-memory methods can become ineffective, with several sophisticated approaches offering little advantage over simple retrieval or even no-memory agents.
By Guannan Lai, Zi-Yang Bai, Jun-Peng Jiang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.