AdaMEM: Test-Time Adaptive Memory for Language Agents

2026-06-04 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

AdaMEM, the Adaptive Memory Agent, is a novel framework addressing the challenge of language agents adapting to dynamic test-time conditions. Existing systems often limit memory retrieval to episode initiation, causing static guidance to misalign during long-horizon tasks. AdaMEM overcomes this rigidity without online model parameter updates by employing a hybrid memory architecture. It maintains a long-term trajectory memory of offline raw experiences and generates dynamic short-term strategy memory on-the-fly to guide decision-making. This mechanism allows for a trade-off between token efficiency and adaptability across different inference-time compute levels. Empirically, AdaMEM significantly outperforms static memory baselines, achieving relative gains of up to 13% on ALFWorld and 11% on WebShop, with consistent leading performance on agentic search for HotpotQA. The framework also introduces STEP-MFT, a Step-wise Memory Fine-Tuning technique, which further enhances adaptation by training the policy to synthesize high-quality strategies from retrieved experiences.

Key takeaway

For Machine Learning Engineers developing language agents for dynamic, long-horizon tasks, AdaMEM presents a critical advancement. Your current static memory approaches likely lead to misaligned guidance over time. You should investigate AdaMEM's hybrid memory architecture, combining long-term trajectory memory with dynamic short-term strategy generation, to significantly improve adaptability and performance. Consider implementing the STEP-MFT technique to further refine strategy synthesis from retrieved experiences, enabling more robust and continuously evolving agents post-deployment.

Key insights

AdaMEM enhances language agent adaptability and performance in dynamic, long-horizon tasks through a hybrid, test-time adaptive memory architecture.

Principles

• Hybrid memory architectures enhance agent adaptability.
• Dynamic strategy generation improves decision-making.
• Continuous reasoning scales agentic memory.

Method

AdaMEM employs a hybrid memory: offline long-term trajectory memory and on-the-fly short-term strategy memory. STEP-MFT fine-tunes the policy to synthesize high-quality strategies from retrieved experiences, enhancing adaptation.

In practice

• Apply to long-horizon tasks like ALFWorld.
• Enhance agentic search performance (e.g., HotpotQA).
• Enable post-deployment agent self-evolution.

Topics

• Language Agents
• Adaptive Memory
• Hybrid Memory Architectures
• Test-Time Adaptation
• STEP-MFT
• Agentic Search

Code references

• yunx-z/AdaMEM

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• AdaMEM: Test-Time Adaptive Memory for Language Agents — Language agents adapt better by dynamically synthesizing context-aware strategies from long-term memory at test-time, rather than relying on static retrieval.
• Memp: Exploring Agent Procedural Memory — Procedural memory, dynamically built and updated, significantly enhances LLM agent performance and efficiency across diverse tasks.
• AI Agents of the Week: Papers You Should Know About — Advancements in AI agents focus on improving memory, planning, collaboration, and reliability through novel architectures and practical tools.
• A 0.12% parameter add-on gives AI agents the working memory RAG can't — Delta-mem offers an efficient, dynamically updated associative memory for AI agents, overcoming RAG and context window limitations.
• Joint Agent Memory and Exploration Learning via Novelty Signals — JAMEL trains agent memory and exploration together via novelty signals, enabling effective, token-efficient exploration in open-ended environments.
• Mem-$π$: Adaptive Memory through Learning When and What to Generate — Mem-$π$ adaptively generates context-specific guidance for LLM agents on demand, outperforming static retrieval methods.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.