FAQ

What is ChunkChef?

A document-processing API: you send files and text prompts, and get back the recognized text and the model's answers for each file. Unpacking archives, conversion, OCR, and the model call all happen on our side.

How is ChunkChef different from plain OCR or a converter?

ChunkChef combines unpacking, conversion, OCR, and the language-model call into a single asynchronous job that runs on your provider key. It's not a format converter, and it's not bare OCR.

What is BYOK, and why is it a good deal?

Bring Your Own Key — you connect your own model provider key. You pay the provider directly for tokens and OCR at their rate; ChunkChef adds no markup.

What am I paying ChunkChef for?

For the tooling: orchestration, reliability (retries, recovery), format and archive support, and ongoing development and support. Tokens and OCR are paid to the provider through your key.

Can I try it for free?

Yes. Sign up, create a key, and send jobs within the free plan — no credit card required.

Can I try ChunkChef without an account?

Yes — the playground lets you upload a file and run a preset prompt, no signup required. It is capped tighter than a real account: a few runs per session, a shared daily budget, and a 10 MB file limit. Create a free account from the result screen to see the full result and keep going without those limits.

What happens to my documents and BYOK keys?

All three secrets (neural.apiKey, ocr.providerKey, and webhookSecret) are never stored in plaintext; the encrypted artifacts and your files are deleted along with the job (~7 days) — see “Security and data” for details.

Which formats are supported?

21 formats: PDF, DOC/DOCX, XLS/XLSX, PPT/PPTX, RTF, TXT, ODT, images (JPG, PNG, TIFF, HEIC), and ZIP/7Z/RAR/EML/MSG archives — see “Supported formats.”

Are there limits?

Yes — on job size, converter response size, timeouts, and prompt size. See “Limits.”

Am I guaranteed valid JSON?

By default, your prompt and the chosen model determine the output structure; there's no server-side validation. Pass `responseSchema` to opt in — ChunkChef then validates the answer and reports `schemaValid`/`schemaErrors` (single-chunk files only).

Which model should I choose?

It depends on the task and budget. Compare intelligence and per-token price across models in the LLM comparison, then set it in your job config.

Can I force OCR instead of hybrid extraction?

Yes — set `extractionMode` to `EXTRACTION_MODE_OCR_ALWAYS` on a BYOK job, or set an account default in the dashboard → Settings. The default, `EXTRACTION_MODE_HYBRID` ("Optimized Hybrid Extraction"), lets the converter choose per file.

Can I get the model to return validated JSON?

Yes — pass `responseSchema` (a self-contained JSON Schema) when creating the job. ChunkChef validates the answer against it and reports `schemaValid`/`schemaErrors` per file (single-chunk files only); a non-conforming answer sets the job to `JOB_STATUS_PARTIAL` rather than rejecting it.

How do I know if I can trust the extracted values?

Set `consensus.mode` (requires `responseSchema`) to have ChunkChef run the extraction 2, 3, or 5 times and vote on each field. You get an agreement score (`minAgreement`) and a list of exactly what the model wavered on — a built-in hallucination check, at k× the token cost on your own key.