Files and recording

What you can drop in, how each source is read, and how you fill gaps later.

An analysis starts with files, not a form. Everything that shows the process goes into one drop: a screen recording, PDFs, photos and screenshots, text files. The model reads them together and writes the documentation from them.

What can go in

SourceFormats and role
Screen recordingVideo (WebM, MP4 and more) — uploaded, or recorded right inside TracAI. It shows the as-is flow and provides the timeline: the strongest evidence. Several recordings of one process can go in together; the longest sets the clock, the others count as observation without timestamps.
DocumentsPDF. Work instructions, manuals, exported boards.
ImagesPNG, JPEG, WebP, HEIC. Screenshots of screens, photos of whiteboards.
TextTXT, Markdown, CSV, HTML. Notes, checklists, exports.
ExcelXLSX. Converted per sheet into table text on upload — sheet names, column letters and row numbers are preserved.
AudioMP3, M4A, WAV, AAC. Meeting recordings — read like a recording, just without the picture.
ArchivesZIP. Unpacked in the browser; the files inside enter the analysis individually, keeping their folder path in the name.

Word files are not read directly yet — attach them as PDF.

Interviewing inside TracAI

If no recording exists yet, record the process directly: share your screen, work through the process once, stop — the recording enters the same analysis as an uploaded file. An AI interviewer comes along: it listens, watches the screen and asks follow-ups exactly where a step would otherwise stay unevidenced. Say what you are doing and why while you click — speech is transcribed and carries the reasoning the picture cannot show.

The interviewer can be switched off: then it is a plain screen recording and nobody asks anything. Open questions in the finished documentation can likewise be answered later by voice — or in chat, if you would rather type.

Recording happens in the browser, so it needs desktop Chrome or Edge; on mobile and in Safari, uploading files stays the way in. You can switch windows while recording — it keeps running.

How the sources are read

A video shows how the process actually runs — observation. A document describes how it is supposed to run — prescription. The documentation keeps the two apart: statements from files are attributed to their source (“according to work-instructions.pdf”), and where video and document disagree, both versions appear in the report with the disagreement as an open question.

Level of detail: file, sheet and column names are carried over exactly — “sheet Woche 27, columns A to C”, not “the sheet”. That is the difference between a note and documentation a stand-in can work with.

Fill gaps instead of starting over

Anything left open becomes an open question in the documentation. Answer it in chat or add another file — say, the video a colleague recorded of part two. The analysis then runs again with the existing document as context, and every state is kept as a version.