About

What we believe about an OCR API done right.

OllamaOCR is a small, opinionated company building one endpoint very carefully. We started it because every team we have ever shipped with — invoicing tools, lending, claims, archives — eventually rebuilt the same tired OCR pipe. We thought it should be a library call. So we wrote it as one.

Four things we will not budge on.

  1. I

    One endpoint. Nothing clever.

    Every product team eventually rebuilds the same OCR plumbing — model picking, retries, page splitting, format coercion. We collapse it into one HTTP call and refuse to add knobs that nobody asked for.

  2. II

    Predictable per-page pricing.

    No tokens, no minutes, no surprise overage tiers buried in a dashboard. A page is a page. You can quote your cost in the meeting where you decide to build the feature.

  3. III

    Clean text, not clever text.

    We return what is on the page — not a paraphrase, not a summary, not a hallucinated table. If a number is smudged, the response says so. We would rather be useful than impressive.

  4. IV

    Boring beats novel.

    Latency we can quote. Errors we documented before you read this. A status page that updates before your customer notices. The pleasure of an API that behaves.

In other words
“We would rather be the OCR endpoint you forget exists than the one you fight with on Tuesdays.”
— internal note, week one

By the numbers

The promises we publish.

847ms

Median latency, single-page invoice.

20MB

File ceiling. PDFs, JPG, PNG, WebP.

$0.0098/page

From, on the Pro plan.

1API call

There is, and only is, the one.

By the byline

Two of us. On purpose.

Small teams ship. We are deliberately keeping it that way until the shape of the company demands otherwise. You will, in practice, talk to one of us.

KA

Karan Avasthi

Co-founder · CEO

I want OCR to feel like Stripe felt the first time you charged a card.
AR

Aral

Co-founder · Engineering

Most of the work is the boring half — retries, file types, the third PDF format you had not heard of.

A counterpoint

What we do not do.

  • We do not sell credits, packs, or surprise add-ons.
  • We do not auto-summarise documents you sent us to transcribe.
  • We do not train on your files.
  • We do not wrap a vendor and call it a model.

A note from the founders

We are, between the two of us, on the receiving end of every support email. If your file fails, one of us will read it. If your bill is wrong, one of us will fix it. If you find this place useful enough to send a note that just says “thanks” — and we hope you do — one of us will reply.

There is a version of this company where we add five SKUs, a reseller program, and a chatbot that calls itself “Ollie.” We are firmly not building that one. We will be honest with you when a feature does not pay rent, and we will keep the API the same shape even when it is unfashionable.

Thank you for reading the about page. They are usually the worst page on the site. We tried.

Karan & Aral

Read enough?

Send a PDF and watch it come back as text. Fifty pages on us.