Xolani: Edit the Transcript, Cut the Media

I keep coming back to one editing problem: spoken-word media is mostly text trapped inside audio and video.

If you are cutting an interview, lecture, podcast, sermon, voice note, or rough talk, the timeline is often the slowest possible interface. You scrub, listen, trim, undo, listen again. The actual decision is usually simple: this sentence stays, that pause goes, this section moves.

Xolani is my attempt to make that loop smaller.

The idea is simple. Import one local audio or video file. Generate a timestamped transcript. Cut the media by deleting words or selecting transcript ranges. Preview the result. Export when it is ready.

The model

Xolani treats the source media as the truth. The transcript sits on top of it.

A word can be corrected without changing its timing. A cut excludes a source-time range, but the original file stays intact. If a cut was a mistake, it can be restored because nothing destructive happened to the media.

The working model looks like this:

source media + transcript words + cuts = composition

That sounds dry, but it matters. It keeps the editor honest. Source time stays stable, exported time is derived from the cuts, and preview can be approximate without pretending to be the final render.

What I am building first

The MVP is deliberately narrow:

import one local audio or video file
validate and inspect it with FFmpeg tools
generate a word-timestamped transcript
make reversible transcript cuts
preview the rough cut
export a finished media file

No collaboration. No cloud workspace. No B-roll suite. Not yet, and maybe not ever.

I want the product to be useful before it becomes broad.

The stack

Xolani is a Rust desktop app built with Iced. The media pipeline leans on FFmpeg and ffprobe. Project state is local, with SurrealDB embedded for persistence. Transcription is behind an adapter seam so the app can support different providers without wiring one vendor through the whole codebase.

That sounds heavier than a prototype, but media apps punish vague boundaries. Files move. Exports fail. Transcripts arrive in odd shapes. The architecture needs boring seams early.

The promise

The product promise is small enough to remember:

Delete words to cut media. Restore any cut. Export when ready.

That is the whole bet. If Xolani can make rough-cut editing feel as direct as editing text, it is worth building.

Follow the build at github.com/lorrehuggan/xolani.