After these pages the service reads a page and returns each word with its box and confidence, on the hosted API or on your own server. This is the whole call, against the hosted API (on your own server: http://127.0.0.1:8080); the key comes on request:
curl -s -X POST "https://api.ocr-software.com/v1/jobs?pipeline=scan-ctc@2&formats=json" \
-H "Authorization: Bearer $OCR_KEY" --data-binary @page.pdf
The order below is the order of a project: install, first call, choice of pipeline, choice of format.
Install the self-hosted OCR server
You download the package, check its SHA-256, install it on your own Linux server with an NVIDIA card and read your first page.
Install →
OCR API guide: first call to production
You go from the first call through key, jobs, retries, limits and errors to a client that survives a bad day.
API guide →
OCR pipelines and pinned versions
You see every fixed, versioned pipeline in one table and pick the one that fits your pages.
Pipelines →
OCR output formats
You learn which of JSON, text, Markdown, hOCR and searchable PDF suits which job.
Output formats →
OCR API reference: every call, field and error
You look up every call, field and error, generated from the OpenAPI file.
Reference →
Writing for an AI agent? It finds the same facts in the agent description.