GrepMem: Plain-Text Log Plus Regex Beats Summarization for Long-Horizon Agents
dair_ai · x · 2026-11-20
USC researchers introduce GrepMem (paper: "Forget Freely, Grep Later"), a lossless, index-free memory mechanism for long-horizon LLM agents: keep the agent's full history as an append-only plain-text log and recall on demand with a regex search tool — no embeddings, retrieval index, or extra model.
Key results (RL-trained Qwen3.5-9B under five context strategies):
- 95-97% on ContextBench vs. at most 21% for lossy baselines
- Beats SUPO summarization by 8.8 points on BCP-Multi
- Stays on the accuracy-vs-token-cost Pareto frontier across all five tasks and scales to longer streams
Why it works: summarization decides what to keep before the agent knows what it'll need later; the log keeps everything and search finds details on demand. A plain-text log plus grep is a simple yet competitive alternative to structured memory systems.
More from coding & agent
- Genie 4 API goes live: precise camera-control for reframing and video — korymath · 2026-11-20
- Tripo Launches Agentic Canvas: AI Plans and Runs Your Full 3D Workflow — LinusEkenstam · 2026-11-20
- OpenAI to Cut Cursor Model Access on Nov. 12 After SpaceX's $60B Anysphere Acquisition — shashib · 2026-11-20
- Infor and AWS Break Down Their Enterprise Agent Stack, With Claude Sonnet 4.6 as Supervisor — shashib · 2026-11-20
- Meta's Benchmark Factory auto-builds coding evals from your own codebase — omarsar0 · 2026-11-20
- Treg Launches /REA-marketing Skill to Reverse-Engineer Any Brand's Playbook Week by Week — cneuralnetwork · 2026-11-20