Deployed LLM agents increasingly rely on a filesystem as their long-term memory: a directory tree of markdown files that the agent itself reads, writes, and reorganizes through generic file tools. Yet this default approach has remained largely unexamined, with prior research focusing on bespoke memory representations and retrieval over them.
In a new paper, researchers at Hugging Face present the first systematic study of filesystem-based agent memory. They formalize the setting as three roles around one memory filesystem: a management agent that integrates and organizes incoming content, a search agent that answers queries with cited sources, and an execution agent whose task trajectories are distilled into skills. This unifies declarative memory and skills in a single store.
Across long-conversation benchmarks and embodied tasks, the team varied memory shape (agent-organized hierarchy, verbatim dump, chunk retrieval), stream scale, tool harness (sandboxed shell, memory-tool-style functions, varied search tooling), and the strengths of the management and search agents, tracking answer quality, cost, and store health as memory grows.
The key finding: organization reliably buys search economy. Organized stores roughly halve retrieval cost where material is large. However, no store shape wins answer quality everywhere, and no agent measured converts organization itself into better answers. The study also reveals that sustainability is a bottleneck—organization erodes for all but the strongest management agent as memory grows—and that changing the tool set alone reshapes the store as strongly as swapping the model.
The paper, titled "Filesystem-Based Memory for LLM Agents: Organization, Evolution, and Sustainability," is available on arXiv (2607.26637). The authors argue the study turns the filesystem default from an assumption into a design space for agent memory.