CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation
arXiv cs.AI · 13 Nov 2026, 04:58 AEST
CoReMem introduces a framework for coding agents that balances memory utility against the costs of writing and reading. This approach enhances task performance by optimizing memory design and actor adaptation.
Key facts
- 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.
- 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.
Read aloud
CoReMem introduces a framework for coding agents that balances memory utility against the costs of writing and reading. This approach enhances task performance by optimizing memory design and actor adaptation. 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. On ChainSWE, CoReMem raises the 27B actor's resolve rate from twenty one point zero percent to twenty five point seven percent, with seventeen percent fewer steps and eleven percent fewer total cache-aware tokens, including memory-module calls, relative to the untrained actor model without memory.
Sources in this cluster
- CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation
arXiv cs.AI · 13 Nov 2026, 04:58 AEST