Feature
File and document handling in ClawAI
How ClawAI ingests files — upload, chunking, OCR, upload checks — and generates them back out as images, documents and research runs with cited sources.
All features · Last reviewed:
Upload, chunking and per-model delivery
ClawAI accepts PDF, DOCX, spreadsheets, CSV, JSON, Markdown, plain text, code files and images. A file is split into passages and indexed, so only the parts relevant to a question are pulled into the prompt, and each model receives the form it handles most reliably — a native image, a native PDF, or extracted text — with every message showing which form each model actually got. Files can be attached per message, including in Compare runs, so several models can be asked about the same document at once.
OCR for scanned documents, and checks on every upload
A scanned PDF with no text layer is run through OCR before it reaches a model, and flagged when recognition confidence is low. Every upload is virus-scanned, checked against its declared file type, screened for dangerous filenames, and rejected if an archive turns out to be a decompression bomb — uploads count against a plan’s file size and storage limits, and files are removed on a retention schedule or deletable at any time.
Generating images, documents and cited research
On the way out, ClawAI can produce an image from a description, export any answer or whole conversation as a formatted file in PDF, DOCX, CSV, HTML, Markdown, TXT or JSON, and run a research task that searches the web, fetches and reads pages, and answers with the sources it actually drew from. Image generation, file generation and research each meter separately (IMAGE, FILE_GENERATION, and the RESEARCH_MODE / WEB_SEARCH / WEB_FETCH / WEB_EXTRACT allowances), distinct from ordinary chat token usage. See how AI tool calling works and what are structured AI outputs, both linked below, for the mechanism behind a defined output shape.
Questions people ask
- What file types can I upload?
- PDF, DOCX, spreadsheets, CSV, JSON, Markdown, plain text, code files and images. Each model receives the form it handles most reliably, and the message shows which form each model actually got.
- Can ClawAI read a scanned document with no text layer?
- Yes — a scanned PDF is run through OCR before it reaches a model, and flagged when the recognition confidence is low.
- What formats can I export a document as?
- PDF, DOCX, CSV, HTML, Markdown, TXT and JSON. Document export is its own metered surface, separate from ordinary chat and from research usage.
Try it rather than take our word for it
Upload, chunking, OCR and upload checks on the way in; image generation, document export and cited research runs on the way out — real, shipped, separately metered features, not one blended file mode.