Private Cloud OCR
The Mathpix API, inside your network
Private Cloud OCR (PCO) is the Mathpix document API packaged as a single container that you run in your own Kubernetes cluster or on a single GPU host, on AWS, GCP, Azure or your own data center. Same endpoints, same models, same outputs as api.mathpix.com. Your documents, page images and results never leave your infrastructure.
Why Private Cloud OCR
For documents that cannot leave your network
Nothing Leaves
OCR runs on your GPU, results land in your bucket, and the only outbound call is an optional license check-in that carries no document content. An air-gapped mode makes no outbound call at all.
The Same API
POST /v3/pdf and POST /v3/text with the same request and response shapes, the same options, streaming and webhooks. Code written against our hosted API works unchanged once it points at your deployment.
Runs Where You Run
One image, one GPU node, Redis and a bucket. Install with Docker Compose on a single host or with the Helm chart on Kubernetes; storage on S3, Google Cloud Storage or Azure Blob.
Built for Whole Collections
A batch jobs API converts a folder in your bucket into a folder of results, with a manifest, a report, retry and cancel.
How It Works
From image to results, on your side
1. Pull the image
Mathpix grants your registry pull access to an image built for your contract, with your license and metering mode inside it.
2. Start it
docker compose up on one GPU host, or helm install on your cluster. Redis and an S3-compatible store are included, or point it at your own.
3. Send documents
Your existing /v3/pdf and /v3/text integration, a batch job over a folder in your bucket.
4. Run it like any service
/health, Prometheus /metrics, structured logs for your logging tool, usage counters you can read yourself, and upgrades that resume documents in flight.
What You Get
Everything the hosted API does
Inputs
PDF, Word, PowerPoint, Excel, EPUB, TIFF and images, with page ranges, hyperlinks, section numbering and every other documented option.
Outputs
Mathpix Markdown and line data for every document, plus Markdown, DOCX, HTML, LaTeX, PPTX and XLSX exports on request, with cropped figures beside them.
Operations
Health and Prometheus metrics, per-day usage on the deployment, signed webhooks, streaming page results, and logs that ship to Datadog, Splunk, Elastic, Loki, CloudWatch or syslog.
When to use it
Private cloud or Mathpix hosted API?
Choose Private Cloud OCR when the documents themselves must not travel: regulated data, customer contracts, or an internal policy. If your constraint is data residency alone, our EU endpoint or the Files API writing results into your own bucket may be the lighter option.
Pricing
An annual license with included usage, on your infrastructure.
From $25,000/year
25 million pages a year included
See plansReady to run Mathpix in your own cloud?
Tell us where you run and what you process, and we will set up a pilot deployment with you.