Ask a Data File a Question Instead of Opening It
Message a Filer folder about a CSV sitting in it and its agent reads the file and answers back in the same chat — the number, not the spreadsheet.
Summary
A folder that piles up CSV or JSON exports — invoices, API dumps, usage logs — answers simple questions the hard way: open the file, find the column, do the arithmetic yourself, every time.
Message the folder's Telegram bot with the question instead. Its agent reads the file that's actually there, works out the answer, and replies straight back in the same chat.
Follow along
Screenshots below are from an actual run against a live Filer instance — a real Telegram message, answered by a real agent reading a real file.
Needs: the Telegram plugin, connected with a bot token and inbound message polling turned on. No extra plugin for a small CSV or JSON file — the agent reads it directly.
Turn on inbound messages, once
Same setup as any other folder that listens for messages — a bot token, a default chat, and "Enable inbound message polling" turned on.
The data file just sits in the folder
No special setup for the file itself — a plain CSV export, the kind that piles up from invoicing, exports, or logging. Nothing has to be prepared or indexed ahead of time.
Ask it something, from Telegram
Send the bot a plain question about the file — no need to open Filer, let alone the spreadsheet. The message wakes the folder's agent, which reads the CSV and works out the answer itself.
"What is the total amount in invoice-aug.csv?"
What it actually replied
"The total amount in invoice-aug.csv is 1,095 (420 + 310 + 275 + 90)."
Telegram isn't the only way in
The point of this walkthrough is asking without opening the spreadsheet, so it leads with Telegram. The same question gets the same answer from Filer's own chat panel at your desk — Telegram is for when you're away, not the only door.
This isn't a mockup — the screenshot above is one real exchange, confirmed against Telegram's own message history: a real message really woke the folder's agent, it really read invoice-aug.csv, and it really did the arithmetic itself (420 + 310 + 275 + 90 = 1,095) rather than repeating a number it was told. What's being verified here is the path — a message from outside the app reaches the right folder, the agent reads the actual file rather than a cached summary, and the reply makes it all the way back out automatically.
Where else this applies
Any folder with data files that pile up faster than anyone reads them fits the same ask-and-answer shape:
- Expense or invoice exports — "what's the total for August" without opening the export.
- API or usage dumps — asking for a count or an outlier from a JSON export without writing a script.
- Handing off to a non-technical stakeholder — someone who'd never open a CSV can still ask it a question over a chat app they already use.
What it needs
- A Filer folder with a trust level that allows message-triggered runs
- The Telegram plugin connected — a bot token, and inbound message polling turned on
- An inference endpoint — Local AI, Frontier AI (API key), or a local LLM server
- For a large file (hundreds of rows or more) or an Excel workbook — the data-analyzer plugin, so the agent reads a structured slice instead of the whole file at once