The short answer
Whizi accepts images, PDFs, Word documents, Excel spreadsheets, PowerPoint decks, and around forty plain text and source code formats. File upload is included on every plan, including Starter, and uploading a file costs no credits.
The four groups, and what happens to each:
| Group | Extensions | What Whizi does |
|---|---|---|
| Images | .png, .jpg, .jpeg | Passed to the model as an image, so the model sees it directly |
| Documents | .pdf, .docx, .doc | Text extracted in your browser, then sent as text |
| Spreadsheets | .xlsx, .xls | Cell contents extracted and sent as structured text; a legacy .xls prompts a re-save as .xlsx |
| Presentations | .pptx, .ppt | Slide text, tables and speaker notes extracted, slide by slide; a legacy .ppt prompts a re-save as .pptx |
| Text and code | 40 plus extensions, listed below | Sent as text as-is |
The complete extension list
Images. .png, .jpg, .jpeg
Documents, spreadsheets and decks. .pdf, .docx, .doc, .xlsx, .xls, .pptx, .ppt. The modern zip-based formats parse in the browser; a legacy binary .xls or .ppt is accepted by the picker but prompts you to re-save it as .xlsx or .pptx and upload again.
Source code. .js, .ts, .jsx, .tsx, .py, .java, .cpp, .c, .cs, .rb, .php, .go, .rs, .swift, .kt, .kts, .scala, .lua, .pl, .pm, .r, .m, .sql, .graphql
Markup and styles. .html, .css, .scss, .sass, .less, .xml, .md
Config and data. .json, .yaml, .yml, .env, .ini, .conf, .csv, .txt
Shell scripts. .sh, .bash, .zsh, .fish
If a format is not on this list, the fastest route is usually to export or save it as PDF, CSV or plain text first. A Google Doc becomes a .docx or a PDF, a Keynote deck becomes a .pptx or a PDF, and almost any data export becomes a .csv.
How extraction actually works
For PDF, Word, Excel and PowerPoint files, the text is extracted in your browser before anything is sent. The model receives the extracted text rather than the original binary file. This has two consequences worth knowing.
The first is that a scanned PDF has no text layer for that extraction to read. Whizi detects an image-heavy PDF from how little text its pages yield and sends the file once through a server-side vision pass instead, which returns a per-page transcript that then travels the same path as extracted text. If that pass cannot produce a transcript, the fallback that always works is uploading the pages as images, since the vision-capable models read a page image directly.
The second is that layout is partly lost. Extraction preserves the words and the reading order, not the visual arrangement, so a complex multi-column report or a heavily formatted table may arrive flatter than it looks on screen. For tables specifically, a .csv or .xlsx upload gives the model far cleaner structure than the same table inside a PDF. What to do with a spreadsheet once it is in is a workflow of its own, set out in how to analyze spreadsheets with AI.
Images are different: they are passed through to the model directly rather than extracted, so charts, diagrams, screenshots and handwriting all work as long as the model you selected can see images.
Retention and deletion
Uploaded attachments and generated media are configured to expire after up to 30 days. You can delete them sooner yourself, and account deletion removes them along with everything else.
Whizi does not use your prompts, files, conversations, voice transcripts, or generated content to train Whizi-owned AI models, and does not sell that content as training data. The full detail is in what Whizi does with your data.
- Images, PDF, Word, Excel, PowerPoint and 40 plus text and code formats are accepted
- File upload is on every plan, including Starter
- Uploading costs no credits
- PDF, Word, Excel and PowerPoint are extracted to text in your browser first
- Scanned PDFs are transcribed page by page through a server-side vision pass
- Tables work far better as .csv or .xlsx than inside a PDF
- Uploads expire after up to 30 days
Frequently asked questions
Can I upload a PDF to Whizi?
Yes, on every plan including Starter, and it costs no credits. The text is extracted in your browser and sent to whichever model you selected, so you can ask a PDF question of GPT, then switch the same conversation to Claude and ask again without re-uploading. A scanned or image-heavy PDF is detected from how little text its pages yield and is transcribed page by page through a server-side vision pass instead.
Can Whizi read Excel files?
Yes for .xlsx. A legacy binary .xls cannot be parsed in the browser, so Whizi asks you to re-save it as .xlsx and upload again. Cell contents are extracted and sent as structured text, which gives the model much cleaner data than the same table pasted into a message or embedded in a PDF. For anything analytical, a spreadsheet upload is the format that produces the most reliable answers.
Is there a file size limit?
The practical constraint is the context window of the model you selected rather than a fixed upload cap. A very long document consumes context that the model then cannot use for anything else, so extremely large files are better split into the sections you actually want analysed. If a document is too long for the model you picked, switching to a larger-context model in the same conversation usually solves it.
Can I upload more than one file at once?
Yes, and this is one of the more useful things to do with a multi-model workspace: upload a spreadsheet and the PDF report it came from together, and ask the model to reconcile them. Drag and drop works anywhere in the chat window, not just on the attach button.
Are my uploaded files used to train AI models?
Whizi does not use your prompts, files, conversations, voice transcripts or generated content to train Whizi-owned AI models, and does not sell that content as training data. Uploads are configured to expire after up to 30 days, and you can delete them sooner or remove everything through account deletion.