Upload / OCR

Try the service. Uses the same API as everyone else, so you need an API key.

API key Engine Mode Language

Expect Pattern Decimals

(result appears here)

API Docs

Base URL: · one endpoint, bearer-key auth.

POST /v1/ocr

Multipart form. Header Authorization: Bearer <key>. Max image 10 MB.

Field	Required	Meaning
`file`	yes	image (png, jpg, webp, …)
`engine`	no	`tesseract` (default) or `rapidocr`
`mode`	no	`no-ai` (default) · `ai-direct` (vision LLM reads the image) · `ai-verify` (engine reads first, LLM checks)
`lang`	no	Tesseract language codes, e.g. `eng`, `chi_sim+eng` (default `eng`)
`hints`	no	JSON string: `{"expect":"digits\|amount\|date\|alnum\|text","pattern":"AAA9999A","decimals":2}`. `pattern`: `A` = letter, `9` = digit, other characters literal. Hints fix look-alike characters (0/O, 1/I, 5/S, 8/B) and layout of amounts and dates.

Response (200)

{
  "text": "RM 12,345.67",
  "confidence": null,        // 0..1, or null when the engine gives none
  "engine": "tesseract",     // or "llm:<model>" when an AI mode answered
  "mode": "no-ai",
  "spans": [],
  "fallback": null,          // "budget" | "no-key" | "llm-error:…" when an AI mode fell back to the plain engine
  "cost_usd": 0.0,
  "ms": 412
}

Errors: 400 bad image/params · 401 missing or invalid key · 413 image too large.

Example

Privacy and limits

no-ai keeps the image on this server. Images are not stored; they are discarded after the response.
ai-direct and ai-verify send the image to a third-party LLM. Check this is acceptable for your documents.
AI use has a monthly spend cap. When it is reached the service answers with the plain engine and sets fallback: "budget".
Accuracy depends on the image. No accuracy figures are promised here; see the benchmark reports for measured numbers.

GET /health

No auth. Returns engine list, whether AI is enabled, and this month's AI spend vs cap.

Status

loading…