Files Om-E understands

Drop a file on a chat, a knowledge base, or the prompt and Om-E turns it into something the model can genuinely read — not a filename it pretends to know. This page defines what "normal use" means, format by format, and is honest about the edges.

Supported formats

Documents: PDF, Word (.docx), PowerPoint (.pptx), Markdown, plain text, CSV, HTML and JSON. PDFs keep their structure — headings, tables, figures — and embedded diagrams are described by a vision model so they come back in search like everything else. Scanned pages are read as images.

Spreadsheets: Excel (.xlsx), including multi-sheet workbooks. Covered in detail below, because sheets carry more meaning than their cells.

Images: PNG, JPEG, GIF and WebP. Each image is described by a vision model on import, and that description is what search finds later — a whiteboard photo surfaces when you ask about the meeting.

Spreadsheets, properly

A working spreadsheet is four layers of meaning, and Om-E reads all four:

Values. Every sheet becomes a clean table with its headers kept close to the data, so a question about row 400 still knows what its columns mean. Dates, percentages and numbers are formatted the way a person would read them.

Formulas.Files saved by Excel carry their computed results and Om-E reads those numbers directly. Files exported by other tools sometimes don't — in that case Om-E shows the model the formula itself, so =SUM(B2:E2) is understood as a row total rather than lost as an empty cell.

Schedules. Project plans drawn as coloured bars — Gantt-style sheets where the timing lives in cell fills, not text — are translated into words: "Build: bars W3 → W5". Ask what happens in week five and the model answers from the plan, not a guess. Background shading and status colouring are recognised for what they are and kept out of the schedule.

Charts.Charts and graphs embedded in a workbook are rendered and described by a vision model — title, axes, legend, and what the bars or lines actually show. The description rides with the sheet, so "what does the sales chart say?" gets a real answer with the exact figures cross-checked from the grid.

What normal use means

The sheets people actually email each other: tables of values with headers, formulas and totals, a few thousand rows, multiple sheets per workbook, coloured project plans, embedded charts. That is the tested, supported path — drop it in and ask questions.

The honest edges

Very large sheets are capped at 5,000 rows each, with a note in the topic when the cap trims anything. Heavily merged headers and pivot tables import as best-effort tables rather than perfect structure. Charts without data labels are described faithfully — the model reads the trend from the picture and the exact numbers from the grid, which is usually the right division of labour. Password-protected files can't be opened at all.

Everything above happens on your machine: extraction, vectoring and the copies themselves stay local, same as every other import. If a real-world file confuses the importer, that's a bug we want — send it through contact.