File upload
Upload files in the Playground to inject their content as context for the current turn. Click to upload or drag and drop.
Supported formats
Text / code: .txt .md .json .csv .log .js .ts .py .sql
PDF: .pdf (text-based only)
Images: .png .jpg .jpeg .webp
Limits
- Each file ≤ 50 MB
- Up to 3 files per message
- Extracted text beyond ~100,000 characters per file is truncated (with a notice), and ~150,000 characters in total per message
- Native media (image / audio / video) sent inline is capped at 15 MB per file — a different, lower ceiling than the 50 MB text-extraction limit
- Calling the API directly, the whole request body must stay under 25 MB; a larger body is rejected by the gateway with a bare 413
About PDFs
- Only text-based PDFs (where text is selectable) are supported
- No scanned-PDF OCR: scanned / no-text-layer PDFs report that text can't be extracted
About images
- With a Vision model selected, the image is sent to the model to "see" (image understanding)
- With a regular model, the image goes through OCR to extract its text as context
- Vision is a beta capability — see Web Search / Vision
Privacy
File contents are not persisted and not logged, and never enter browser local storage — they are used only as context for the current request. Uploads are scanned for secrets (private keys etc. are blocked).
Common errors
| Symptom | Meaning |
|---|---|
| File too large | Over 50 MB |
| Unsupported format | Not in the list above |
| PDF has no text layer | Scanned PDF, text can't be extracted |
| Secret detected | File contains a suspected key and was blocked |
| File corrupted | Could not be parsed |