frisket DocsSearch

Choose an approach

Engines and models

An action is the task you ask Frisket to do. The engine is what does it: a program that runs with Frisket, or a model at a hosted provider.

Listen where the words differ

Sourcefaster_whisper · baseMAI Transcribe 2
Policy Committee · opening 30 secondsShe's out there todayShe's out ill today

These are actual outputs for the same opening 30 seconds. The engines disagree on this phrase. Listen to that passage before using or correcting it.

Choose within the action

Many actions can be done more than one way. OCR can use RapidOCR, Tesseract or a vision model; Extract fields can use dozens of models. The choice is called the engine. Engines differ in three ways that matter:

To choose, click the engine box in the action's panel. The picker groups engines by where they run. Click one to see its details: where it runs, what it needs, and what it costs.

The OCR engine picker

Set up an unavailable choice

An engine marked Setup needed needs something before it can run. Its details say what:

Check what leaves your device

Each engine is labelled with where it runs:

If your documents are confidential, or you've promised a source they won't leave your control, choose an engine that runs with Frisket, and check the label before every run with a new engine.

Try the same source

Before running an engine on everything, try it on a few rows and check the results against the source. When comparing engines, use the same rows and settings for each, so the engine is the only difference.

Every run shows an estimated cost first. Runs up to your approval limit go ahead without another prompt; above it, Frisket asks you to confirm. Set the limit under Settings › Personal › Profile, in Cost pre-approval amount (USD).

Compare OCR

Compare transcription

Compare action variants

Choose an OCR engine

Choose an extraction model