filesystem-context

To manage unlimited context, you can offload large tool outputs to the filesystem and use search tools like grep to retrieve specific information dynamically, preventing context window bloat.

Filesystem-based agent memory works by writing plans, learned preferences, and sub-agent communications to files, allowing the agent to retrieve and reload this persistent state on demand.

Yes, you can use file search tools like grep and glob for dynamic context discovery, retrieving specific information from the filesystem on demand rather than loading everything into the context window at once.

The best way to store large web search results is to offload the raw HTML or tool output to a file, then use grep to extract only the relevant details, keeping the agent's context window clean and focused.

Yes, this approach supports scratch pads, plan persistence, and sub-agent communication by leveraging file operations to save and retrieve intermediate results and persistent state across the filesystem.

You should avoid filesystem context management if your agent environment lacks file operation capabilities or if your task requires keeping all data strictly in memory without filesystem persistence.