Upload file sources
Add supported documents and images to an agent, check extraction status, and handle unreadable files.
Fetchply accepts PDF, DOCX, PPTX, CSV, XLS/XLSX, TXT, Markdown, HTML, JSON, PNG, JPEG, WebP, and GIF knowledge files, each up to 10 MB; your total stored training data also depends on your plan.
- Documents (PDF, DOCX, TXT, Markdown, HTML): the readable text is extracted and indexed. Images inside documents are not read.
- Presentations (PPTX): the text on each slide and its speaker notes are indexed. Images, charts, and layout are not read.
- Spreadsheets (CSV, XLS/XLSX): rows and values are indexed so the agent can answer questions about them.
- Data files (JSON): the document is indexed as structured data, useful for product feeds or configuration-style content.
- Images (PNG, JPEG, WebP, GIF): the configured vision model reads visible text and useful visual details, then indexes them as structured knowledge. Clear, high-resolution images work best.
Older Word (.doc) and PowerPoint (.ppt) files are not supported; save them as DOCX or PPTX first.
Prepare the file
Remove passwords, private data, duplicate pages, stale policies, headers that overwhelm the content, and unsupported media-only pages. Crop standalone images to the useful content and make sure important text is clear.
Upload the file
Open Training Data → Files, choose the file, and keep the page open until the upload completes.
Wait for indexing
Processing means extraction or embedding is still running. Indexed means the source is available for retrieval. Failed needs attention.
Preview and test
Preview the extracted knowledge, then ask a question that depends on a specific part of the file.
Scanned PDFs need searchable text before upload. Fetchply does not promise OCR for image-only pages.
Files with almost no extracted text can fail. Convert scans to searchable PDFs or paste the approved content as a text source.
Troubleshooting
- "Unsupported file type": check the file extension against the list above. Older .doc and .ppt files need to be saved as DOCX or PPTX before uploading.
- An HTML file indexed less content than expected: navigation menus, scripts, and styling are removed on purpose so only the readable content is trained.
- A JSON file failed to process: the file must be valid JSON. Open it in a text editor and fix any syntax errors, then upload it again.
- A presentation shows very little text: only slide text and speaker notes are read. Slides that are mostly images or charts contribute little content.
- An image failed to process: confirm it is a PNG, JPEG, WebP, or GIF under 10 MB. Try a clearer image with readable text. If image reading is unavailable, ask your administrator to check that a vision model is active.
- An image indexed incomplete information: open its preview to see exactly what was extracted. Crop away unrelated areas or upload separate close-ups when small labels or dense tables are difficult to read.