Cost-aware memory improves coding agent success and efficiency
CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation
Artificial IntelligenceComputation and LanguageSoftware Engineering
Summary
When computer programs work through a series of coding tasks, figuring out what past information to remember is tricky because future needs are unknown. The authors introduce a method called CoReMem that helps these programs pick useful memory pieces while considering the cost of storing and using them. It also guides the program to adapt its behavior to use this memory efficiently for upcoming tasks. Tested on a coding evaluation, CoReMem improved success rates and made the programs faster and less resource-intensive.
What this means in practice
- •For software development teams: Enhance automated coding assistants to retain and use evolving code context efficiently, improving task success and reducing computing resources.
- •For cloud platform operators: Optimize caching and memory strategies for automated code generation services to lower costs while maintaining output quality.
Authors
Peiwen Li, Harshit Verma, Sophia Zhou, Haiwen Wang, Ngoc Bui, Heng Chang, Fei Fei, David van Dijk, Rex Ying
Abstract
Coding agents tackle successive tasks on an evolving repository, carrying forward edits and writing memory before the next task is known. Selecting memory for unknown next tasks requires balancing future usefulness against the costs of writing and reading it. The actor must also decide when acting on available memory saves effort as the task unfolds. We introduce Coding-Repository Memory (CoReMem), a cost-internalizing framework that guides memory design and actor adaptation through measured memory value on the next task. Prospective segment valuation compares runs of the next task with and without individual trajectory segments from the same inherited repository state, guiding a compact memory built around anchors: repository locations and executable checks. Actor adaptation estimates on-policy memory value within rollout groups with and without memory, using a reward that prioritizes resolution and penalizes step count in successful runs. Provenance-gated value credit assigns the additional value term to steps identified as using memory-supplied anchors, while every step retains outcome credit. On ChainSWE, CoReMem raises the 27B actor's resolve rate from 21.0% to 25.7%, with 17% fewer steps and 11% fewer total cache-aware tokens, including memory-module calls, relative to the untrained actor model without memory.