Vision reads documents and images and answers your question about them.
The default model is PailotOCR, Pailot's document recognition that runs inside the company contour:
- Takes a file or a URL: a PDF or a PNG, JPEG or WebP image, uploaded or passed as a URL such as
<start.pdf>. A platform file is read only when you can access it, and an external URL is downloaded with protection against internal network addresses. - Recognises the whole document: text of every page, tables, headers, footers and page numbers.
- Answers with pailotgpt 3: the local model answers your prompt from the recognised text.
- Needs no API key: the document is not sent to external services.
- Returns the full document:
contentis the answer,documentis the whole recognised document for the next blocks andwarningslists warnings.
Other models:
- pailotgpt 3: the local multimodal model that describes photos, diagrams and videos.
- gpt-4o and Claude 3: cloud models that need the provider's API key.
Vision does not generate images; it only reads them.
Usage Instructions
Integrate Vision into the workflow. PailotOCR, the default, recognises a whole PDF or image inside the Pailot contour, answers the prompt with pailotgpt 3 and returns the full recognised document. Other models analyze images directly.
Which file to pass
- A document the workflow keeps using: upload it once in Settings → Files, then open the block's "Image File" field and pick it from the list — every workspace file is listed there next to "Upload New File". No URL and no storage key are involved. The field is available in both block modes.
- One-off run: upload the file into a test run or send it in a chat. That file belongs to its run — upload it again for the next run.
- When you do need a URL (to use the file in another block): press "Copy link" next to the file in Settings → Files. Never write a URL or a storage key by hand — the key is assigned at upload time and cannot be derived from the file name.
- If the block reports "file not found": the file behind that URL is no longer in storage. Upload the document again and pick it in the block.
- Where the reference goes: in basic mode use "Image File" or "Image URL". "Image File Reference" only works in advanced mode; a reference left there while the block is in basic mode is dropped and the block answers "Either imageUrl or imageFile is required".
PailotOCR warnings and errors
| Code | Meaning | What to do |
|---|---|---|
document_incomplete | Some pages were recognised only partially; the message lists them | Check those pages in document or upload a better scan |
document_truncated_for_answer | The document is longer than the pailotgpt 3 window, so the answer uses the start of the text | The full text is still in document; split long documents or use a knowledge base |
When the PailotOCR service is not configured or unavailable the block returns an error and does not send the document to a cloud model. An unsupported file, such as a video, gets an error that says why. When a URL points to a file that was already removed from storage, such as a test file from an old run, or one you cannot access, the block reports "file not found"; upload the file again.
Tools
vision_tool
Read a document with PailotOCR or analyze an image with a vision model.
Input
| Parameter | Type | Required | Description |
|---|---|---|---|
apiKey | string | For cloud models | API key for the selected provider; not needed for PailotOCR or pailotgpt 3 |
imageUrl | string | No | Image or document URL: public, or a platform file from a previous block (<start.pdf>) |
imageFile | file | No | Uploaded file: PDF, PNG, JPEG or WebP for PailotOCR; images and videos for other models |
model | string | No | pailot-ocr (PailotOCR, default), qwen3.5:9b (pailotgpt 3), gpt-4o, claude-3-opus-20240229, etc |
prompt | string | No | Question or task about the document or image |
Output
| Parameter | Type | Description |
|---|---|---|
content | string | The answer or description |
model | string | The model that was used (PailotOCR for documents) |
tokens | number | Total tokens used for the answer |
usage | object | Detailed token usage breakdown |
↳ input_tokens | number | Tokens used for input processing |
↳ output_tokens | number | Tokens used for response generation |
↳ total_tokens | number | Total tokens consumed |
document | json | PailotOCR only: the full recognised document with pages, blocks in reading order, table cells and processing status |
warnings | array | PailotOCR only: document_incomplete and document_truncated_for_answer warnings |