PRODUCT · PROCESSORS
Composable units of work. Mix to fit your pipeline.
Use each processor on its own from the API, or stack them into a pipeline.
Pick a processor. Read the deep-dive.
Every processor has a focused job and a published API. Stack them and you get a pipeline; run them alone and you get a single endpoint.
Convert
PDFs, Office docs, spreadsheets, images → markdown, HTML, JSON, or pre-chunked output.
- PDF · DOCX · PPTX · XLSX · EPUB · HTML
- 90+ languages of OCR
- 89.9% on olmOCR-bench tables
- Parse Quality Score on every call
Extract
Schema-defined structured data — typed JSON with a citation block ID on every field.
- JSON Schema · Pydantic
- Block-ID citations per field
- Enum normalization built-in
- Arrays for long line-item tables
Segment
One multi-document PDF in. Many named documents out.
- Auto-detect document boundaries
- Each segment named + paged
- Confidence per segment
- First step in a multi-doc pipeline
Customize
Describe a transform in plain English. The assistant builds the processor.
- Chat-driven, AI-generated
- Modifies Convert's output
- Reusable by processor ID
- Composes inside Pipelines
Eval
Rubrics across a reference corpus, scored 0–5. Gate pipeline promotions on a number.
- 0–5 rubric scoring
- Reference-corpus diff between runs
- Pin a rubric to a pipeline version
- Score gates the next promotion
START
Five processors. One platform.
Free tier, no credit card. Start with Convert and add the rest as you need them.