Filer AI logoFiler AI

Filer Usecase

Catch the Value That Doesn't Belong

A Filer folder agent notices when a metrics file updates and points out the one value that clearly stands out from the rest — no dashboard to open, no graph to eyeball.

  • File event
  • Chat

What you get

A file of response times, error counts, or any other repeating measurement is only useful if someone actually looks at it. Most of the time nobody does — until the number that mattered has already scrolled off the bottom.

Point a Filer folder at the file instead. Its agent wakes up every time the file changes, reads the new value against the recent ones, and says plainly whether it looks unusual — in the same chat you'd use to ask it anything else about the folder.

How to do it

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

Needs: a trust level that lets the agent run on file changes and read the folder. No plugin required.

What it needs

  • A Filer folder with a trust level that allows file-triggered runs
  • No plugin, no external account — this one runs on Filer's own built-in file-reading tools
  • An inference endpoint — Local AI, Frontier AI (API key), or a local LLM server

Tip: a trust level that only allows reading (no file creation or editing) is enough for this case — the agent never needs to write anything back to answer "does this look unusual."

  1. A metrics file, nine unremarkable rows

    A short instructions file already in place tells the agent what to do whenever response-times.csv changes: read it, compare the newest value against the others, and say whether it stands out — or say plainly that nothing looks unusual.

  2. A new row lands, the agent flags it on its own

    Appending one line — a response time of 850ms next to nine values sitting between 97 and 105ms — is the trigger. Nobody asked; the file changing is enough to wake the agent.

    What it actually replied
    "...850ms...approximately 8 times higher [than the 97–105ms range seen in the rest of the file]...an anomalous outlier."

What this reliably catches:

a value that's obviously off, the way 850ms is obviously off next to a run of 97–105ms values. A value that's only mildly elevated — say, 40–50% above the recent range instead of 8x — is a judgment call the agent doesn't reliably complete every time. This is a scope Filer's own testing found, not a claim to take on faith: treat this as a first-pass flag for clear outliers, not a statistical anomaly detector for borderline drift.

Verified

This isn't a mockup — the screenshots above are from one real run: a real file change really woke the agent, and the value it called out (850ms against a 97–105ms baseline) really is the value that was added, with the correct comparison spelled out rather than just a bare "yes." A second, separate run with the same file updated to another ordinary value (no outlier) really was triggered too, and correctly replied "nothing looks unusual — the values are consistently within a tight range of 97–105ms" — not pictured here since there's nothing to show for a negative result, but it's why this case doesn't just claim detection, it claims detection of clear outliers specifically (see the scope note above).

Where else this applies

The mechanism is generic — watch a file, compare the newest entry to the recent ones, say what's off. Any repeating log or metric nobody has time to babysit is a fit:

  • Error-count logs — a build or job that logs its failure count per run, flagged the moment one run's count is far outside the usual range.
  • Daily counts — orders, signups, tickets — any per-day tally where "way more or way fewer than usual" is worth a second look.
  • Combine with a messenger trigger — pair this with the same folder's Telegram or Slack connection so a clear outlier reaches you as a message instead of waiting to be read in Filer's own Activity view.

Other ways in

Right-click
No right-click action yet — use chat.
Chat
The walkthrough above shows the trigger path — the file changes, the agent notices on its own, nobody asks. The same check is one chat message away too: open the folder's chat and ask something like "does anything in response-times.csv look unusual?" and the agent reads the same file and answers on demand instead of waiting for a change. There's no dedicated right-click action for this one yet — chat is the manual path until there is.
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