Agent memory stops only looking back: Pro-Memory splits future intentions from the moment to act
Synopsis
Most memory systems for large language model agents only store and retrieve past information, while proactive agents also need to hold future intentions and realize them when the right moment arrives. Pro-Memory separates keeping a future intention from deciding when and how to realize it: pending objectives stay linked to their supporting evidence, new observations are checked against the original goal and activation conditions, and contributions are coordinated with the agent's ongoing actions. On two benchmarks covering delayed-intention tasks and open-ended proactive assistance, proactive task performance improves consistently over representative memory approaches across multiple LLM backbones while unnecessary interventions are limited.
Interpretation
Proactive large language model agents gain a memory route dedicated to future intentions rather than only retrieval of past information. Earlier agent memory systems are retrospective, preserving and retrieving past information to inform current decisions, and leave no place for intentions about what to do later. On two benchmarks spanning delayed-intention tasks and open-ended proactive assistance, proactive task performance improves consistently over representative memory approaches across multiple LLM backbones, while unnecessary interventions are limited.
The system handles remembering what should be done and judging whether now is the time as two separate steps. Pro-Memory is a memory framework inspired by cognitive theories of human prospective memory that separates maintaining future intentions from deciding when and how to realize them. It preserves pending objectives linked to their supporting evidence, assesses new observations against the original goals and activation conditions, and coordinates with the agent's ongoing actions; intention states update from actions actually issued or evidence of resolution, and an experience memory distills feedback from completed episodes into guidance for later decisions.
Perspective
The demonstrated scope is delayed-intention tasks and open-ended proactive assistance on two benchmarks, compared with other representative memory approaches across multiple LLM backbones; the material is summary-level and reports no specific metric values.
The material states only that proactive task performance improves consistently and that unnecessary interventions are limited; it gives no size of the gains and no differences among backbones.