Give agents a read-only filesystem over your docs

With Agno's PageFileSystem toolkit, an agent works through your published pages with ls, tree, find, cat, head, tail, wc, rg and grep, the way an engineer works through a repo.

A user asks your docs agent about page_store. Vector search returns a chunk that sounds close, and the agent answers from that chunk. An engineer would have grepped for the exact name and opened the page it lives on. With PageFileSystem, the agent can do the same.

from agno.agent import Agent
from agno.knowledge.page import PageFileSystem
 
page_files = PageFileSystem(knowledge=knowledge)
agent = Agent(tools=[page_files.tools()])

Vector search still finds the right page, and the agent then reads all of it.

What can't the agent do?

PageFileSystem can't run a shell, expand shell expressions or write files. Reads use pinned page revisions, listings return scoped metadata, and grep is a bounded literal scan. Output stops at 30,000 characters with a continuation notice, and Agno caps regex time, command time and read volume.

The toolkit works in sync and async runs, and you can rename the tool with tools(tool_name=...).

See the cookbook, and learn more about published pages in the documentation.