Sprint · pixels to data

Read what changed.
Remember the rest.

A fast reader for familiar screens. Select a value and its label once. Read new pixels, reuse unchanged regions, and leave uncertain values empty.

First read

—

Includes worker startup if cold

Unchanged screen

—

Image decode + exact pixel checks

Hosted model calls

—

Recognition runs locally on the server

Scenario checks

—

Measured after you run

The observation bench

Synthetic screenshot · 960 × 520
Synthetic store screenshot: baseline

Green regions contain the values. The reader also checks each label before accepting a number or status.

Run the benchmark to inspect six real extraction results.

Structured output

Read the image into typed values, with a source for every field.

{
  "revenue": null,
  "orders": null,
  "active": null
}
  • RevenueAwaiting frame
  • OrdersAwaiting frame
  • Store activeAwaiting frame

01

Check the pixels

Exact comparisons of the value and label. A change elsewhere can be ignored.

02

Read the changes

Local OCR reads just the new regions. Strict parsers turn text into typed values.

03

Keep the evidence

Every value records its source. Missing text and changed labels leave empty fields.