filesystem-context

Persistent agent memory uses the filesystem to store large tool outputs and scratch work. This reduces prompt size while keeping data accessible on demand through lightweight file references, preventing context bloat during complex tasks.

Plan persistence saves long-horizon workflow states to the filesystem, allowing you to store and rehydrate plans across sessions. This ensures continuity without losing intermediate progress during multi-agent coordination.

Create a scratch pad by writing intermediate results and large tool outputs to files on the filesystem. Retrieve only necessary slices or lightweight file references during the conversation to maintain a minimal active prompt.

Yes, filesystem context supports sub-agent workspaces by providing isolated collaboration areas. Sub-agents write findings to files, enabling coordinated synthesis without overwhelming the main agent's active prompt.

Use filesystem-based context management for long-running multi-agent workflows, processing lengthy datasets, or dynamic context discovery. It prevents prompt bloat when large tool outputs must remain accessible without consuming active context window space.