Filer AI logoFiler AI

Filer Usecase

Find the Near-Duplicates Before You Do

A Filer folder agent reads through a folder's files and points out which ones are near-duplicates of each other — reworded drafts, repeated exports — without changing anything itself.

  • File event
  • Right-click
  • Chat

What you get

Drafts, downloads, and scanned copies pile up in the same folder over time, and some of them are really the same content with different wording or a "-draft" suffix. Figuring out which is which means opening each one and comparing by eye.

Point a Filer folder at the pile instead. Its agent reads the folder's files, compares what's actually in them, and tells you which ones look like near-duplicates — leaving the decision about what to keep or delete to you.

How to do it

Screenshots below are from an actual run against a live Filer instance, using placeholder notes in place of real documents.

Needs: the multi-file-tools plugin, installed from Connect.

What it needs

  • A Filer folder with a trust level that allows file- or schedule-triggered runs
  • The multi-file-tools plugin, installed from Connect (reads several files in one batch)
  • An inference endpoint — Local AI, Frontier AI (API key), or a local LLM server

  1. Three files, two of them alike

    Two planning memos say the same thing in different words, and one unrelated file sits alongside them. A short instructions file tells the agent to read the folder's text files in a batch and report near-duplicates — read-only, nothing gets created or changed.

  2. It reads them together and calls out the match

    A file change is the trigger. The agent reads the folder's files together, compares their content, and correctly names the two reworded memos as near-duplicates while leaving the unrelated file out of the call — no file created or modified in the process.

    What the agent actually wrote back
    "memo-v1.txt: Contains 'Q3 Planning Memo' with specific priorities: expand onboarding funnel, ship mobile beta, reduce support ticket backlog by 20%. memo-v2-draft.txt: Contains '(draft)' prefix and similar content but rephrased as 'growing the onboarding funnel' instead of 'expand,' 'shipping the mobile beta' instead of 'ship,' and 'cutting...by twenty percent' instead of 'reduce...by 20%.' recipe.txt: Completely different file, unrelated to the memo content. — memo-v1.txt and memo-v2-draft.txt are near-duplicates because they contain nearly identical messaging about Q3 priorities (only minor wording changes...). They likely represent the same content with slight variations."

Verified

This isn't a mockup — the screenshot above is from one real run: the agent really read all three files, really identified the two reworded memos as near-duplicates while correctly leaving the unrelated file out, and really created no new file — confirmed by checking the folder's contents afterward, not just the agent's own report. Comparing wording is regular language-model reasoning; what's actually being verified here is that reading several files together and reporting back, read-only, works end to end.

Where else this applies

The mechanism generalizes to any pile of text-based files where the same content might exist more than once under different names — not just planning memos:

  • Downloads folder — the same file saved twice under a "(1)" suffix, or re-downloaded weeks apart.
  • Email attachment archive — the same attachment saved separately from several threads it was forwarded through.
  • Shared draft folder — several people's independent drafts of the same document, before anyone realizes they overlap.
  • On a schedule instead — a folder that gets too many small changes to trigger a comparison on every one of them can instead sweep the whole folder for near-duplicates once a week on a schedule, rather than reacting to individual file events.
  • Notify instead of reading in Filer — pair this with the same folder's Telegram or Slack connection so the near-duplicate call reaches you as a message instead of waiting to be read in Filer's own Activity view.

Other ways in

This is what the right-click path looks like on the same two files:

Right-click
Chat
The walkthrough above shows a file-change trigger running the comparison on its own. The same batch-read comparison doesn't need a trigger at all: select files — or right-click a whole folder — and choose Find Duplicates from the multi-file-tools plugin, or just ask in chat, "find near-duplicates in this folder." All three paths call the same underlying batch file-read; the trigger is only how this particular run started.
Other automation triggers
Not shown in this walkthrough.
Other integrations
Not shown in this walkthrough.
See all usecases

Try it in your own folder

Download Filer