Inside the app

Quick to record. Slow to write up. Not anymore

One pass turns the recording into a timestamped transcript, numbered steps, pictures pulled from the moments they belong to, and an export your own system can read.

The transcript comes back timestamped

Your video file stays where it is. The sound of the recording is what gets transcribed, and what comes back has a timestamp on every line. Those timestamps are the whole trick: they are what lets a picture be pulled from the moment a thing was described. The transcript is then never edited, because it is the record of what was said — corrections happen in the article, so you can always check the draft against the source.

00:03:58 so now click into settings, and you will see

00:04:12 the export panel down on the left hand side

00:04:26 pick markdown if you are using a static site

Every step stays tied to its part of the recording

Each section carries the span of footage it came from. Insert a picture into step five an hour later and the frame picker opens at step five, not at the beginning of a forty minute file.

Step 5, choose an export format 04:12 to 05:40

Pull a frame, then mark it up

The frame picker lands on the moment you asked for. From there it is a real editor: crop the frame, box and arrow the part that matters, set type on it in a typeface you imported yourself. Nothing is flattened until you export, so re-picking a frame or moving a box a week later costs you nothing else.

Editing made easy, and do not overthink the design

You are editing the finished page the whole time; there is no preview mode to toggle. Do not spend the hour on how it looks here, though, because wherever this lands will have a design of its own. What matters is that the export carries everything your system needs to lay it out itself.

Markdown with a folder of pictures .md + /images
HTML with HowTo schema .html
A manifest and README written for an agent manifest.json
Straight to WordPress or Webflow connected

The AI behind it

Three places the AI can run, and you choose per job

Transcribing and drafting are the only two things that need a model at all. Everything after them — editing, re-picking a frame, marking up a picture, exporting — happens on your computer whichever of these you pick, and costs nothing.

The built-in AI service

Nothing to set up: the models are chosen and kept current for you, and each job says what it will cost before it runs. This is the only path credit pays for.

Your video file stays
The sound of the recording, to extract the transcript sent
The transcript text, to write the draft sent

Paid for with credit

Models on your own computer

Point the writing at Ollama or at llama.cpp, and the transcript at the speech engine that ships with the application. Both run on your own hardware, and the application will work with no network at all.

Your video file stays
The sound of the recording stays
The transcript text stays

No credit and no API bill

Your own API key

A console key with billing on it, held on your machine and used directly. The work goes to your own account with that provider, so your usage and your invoice are theirs and ours to neither.

Your video file stays
Whatever that provider needs, to your account sent

Billed by your provider, never by us

Settings lists every service you have set up and says what each one sends, and no job runs until you have seen that. Nothing is chosen for you and nothing is a default you did not set.

Your computer

What the speech engine runs on

The speech engine is the local half of the second option above. It extracts the transcript on your graphics card or on your processor, and Settings offers the choice whenever a usable card is found.

Graphics card

Any NVIDIA, AMD, or Intel card whose driver supports Vulkan, the graphics interface the engine is built on. Vulkan comes with the driver rather than the card, and the current driver from any of the three makers includes it for cards going back roughly a decade.

If Settings reports no graphics card on a machine that has one, install the latest driver from the card’s maker. Virtual machines, remote desktop sessions, and Windows’ basic display adapter have no Vulkan driver, and on those the engine uses the processor.

Processor

Any 64-bit Intel or AMD processor. The engine ships one build of its processor code per instruction set, from a plain baseline up to AVX-512, and loads the best one your machine supports.

There is no Windows on ARM build.

Before anything is sent

Every job tells you what it is about to send, and what it will cost, before it runs

That is the whole promise: not that nothing ever leaves your computer, which would be false the moment you chose a cloud service, but that you are never the last to know.

Start your trial See pricing

7 days, no card.