Document Parse
Convert a PDF or image to clean Markdown with the gateway's LLM-based OCR endpoint — playground, API contract, and code samples.
Document Parse turns a PDF or image into Markdown using a vision-capable LLM of your choice. The built-in extraction prompt preserves document structure — headings, tables, lists, emphasis — skips headers and footers, and returns an empty string for blank pages. Try it in the playground, then call the same pipeline at POST /v1/ocr/parse.
Parse a document in the UI
Open the playground
In your project, open AI Gateway and switch to the DOCUMENT PARSE tab. The PLAYGROUND sub-tab is active by default; the CODE sub-tab shows an equivalent API snippet.
Choose the LLM
Use the Select LLM dropdown to pick a provider, then choose the model and the Connection holding that provider's credentials. The Settings gear exposes temperature, max tokens, and — for models that support it — reasoning effort. Pick a vision-capable model: it will read page images directly.
Upload and parse
Drop a PDF or image file into the dropzone and click Parse File. When parsing finishes, the extracted Markdown appears next to the file preview.
A parsed document showing the original file preview beside the extracted Markdown text
screenshot: gateway-parse-result
How files are handled
The endpoint detects the file type from the filename's MIME type, falling back to content sniffing (a %PDF header means PDF; anything else is treated as an image). PDFs run through the LLM PDF Converter and images through the LLM Image Converter — the same nodes you can use inside workflows — with one output document produced per file.
Call the API
POST https://api.getdynamiq.ai/v1/ocr/parse is a multipart/form-data request with two fields:
filefilerequiredoptionsstring (JSON)requiredThe options JSON:
llmobjectrequiredstreambooleanimport json
import os
import requests
response = requests.post(
"https://api.getdynamiq.ai/v1/ocr/parse",
headers={"Authorization": f"Bearer {os.getenv('DYNAMIQ_ACCESS_KEY')}"},
files={"file": open("document.pdf", "rb")},
data={
"options": json.dumps(
{
"llm": {
"type": "openai",
"model": "gpt-4o",
"connection_id": os.getenv("DYNAMIQ_CONNECTION_ID"),
"temperature": 0.0,
"max_tokens": 4096,
},
"stream": False,
}
)
},
)
response.raise_for_status()
print(response.json()["data"]["text"])curl https://api.getdynamiq.ai/v1/ocr/parse \
-H "Authorization: Bearer $DYNAMIQ_ACCESS_KEY" \
-F "file=@document.pdf" \
-F 'options={"llm": {"type": "openai", "model": "gpt-4o", "connection_id": "'"$DYNAMIQ_CONNECTION_ID"'", "temperature": 0.0, "max_tokens": 4096}, "stream": false}'options must be a JSON string inside the multipart form — serialize it with json.dumps rather than passing a nested object.
Response
{
"data": {
"text": "# Invoice 4812\n\n| Item | Amount |\n| --- | --- |\n| Pro plan (June) | $49.00 |"
}
}data.text is the trimmed Markdown content of the document. With "stream": true the response is an SSE stream of extraction events instead.
Errors
| Status | Cause |
|---|---|
422 | The options field is not valid JSON or fails validation |
400 | The OCR run failed — unreadable file, LLM error, or no text extracted; the error detail explains which |
401 | Missing or invalid credentials |