Files
Upload a tender document, have its text read out, and ask the AI what it actually says.
Files is where documents live: a request for tender, a statement of requirement, a previous response, your own draft. Upload one and its text is extracted server-side, which is what makes the next part possible — you can ask the AI what the evaluation criteria weight, whether a mandatory condition has been met, or what the response template actually demands, and get an answer that quotes the document rather than summarising the idea of one.
Everything here is your organisation's own material. Everyone in your team can open a file and its text, it is never used to train a model, and deleting a file deletes both its text and the file itself. What the AI is sent, and when, is on AI and your data.
Uploading
Press Upload a document and pick a file. Four kinds are read:
| Kind | What is read |
| --- | --- |
| PDF (.pdf) | The text layer, page by page. A scan with no text layer is refused with that reason — see below. |
| Word (.docx) | Paragraphs and table cells, in document order. |
| Text (.txt) | The file as it stands. |
| Markdown (.md) | The file as it stands. |
A file may be up to 20 MB, and your organisation may hold 50 of them at once. Both limits are checked before the upload starts and again by the server, so a file that is too large is refused in a moment rather than after a long upload.
The upload goes straight from your browser to your own storage folder — the file itself never passes through this app. What the server is given is a short-lived link to read it once, which is how the text is extracted; the link expires in ten minutes and is not kept.
While it is being read
A new file shows as Reading… and the page refreshes itself until it is done. Extraction is a parser rather than a model, so it is usually seconds; the work finishes whether the tab stays open or not.
Then the row says one of:
- Ready, with what the document turned out to hold — pages for a PDF, and the number of characters of text. A document longer than 400,000 characters is kept to its first 400,000 and says so; the front of a tender is where the requirement lives.
- Failed, with the reason in plain words. The common one is no text layer: a PDF that is a photograph of a page has no text to extract, and this step does not run OCR on it. That is said out loud rather than shown as an empty document. Re-save or export the file as a text PDF and upload it again.
Asking about a document
Ask AI on a row opens AI research with that document attached. In the conversation itself, Attach (the paper clip) above the composer attaches any of your ready documents — up to five on one question.
The agent reads an attached document in slices, so a 150-page tender is read in parts rather than pushed whole into one answer, and it tells you which file and which part a quotation came from. It treats the contents as evidence and never as instructions: a document that contains something looking like an order to the agent is flagged rather than followed.
Preview and delete
Preview shows the first 20,000 characters of the extracted text — which is what the AI reads, so it is also how you check that a document came out readable. Layout, images and signatures are not part of it.
Delete removes the row and the stored file together. A conversation that used the document keeps the name it was asked with, because that is the record of what an answer was built from.
Not yet
- Attaching a document to a pursuit — not built yet.
- Automated capture of AusTender's own tender documents — they sit behind a login, and their terms need reading before anything automated fetches them. Listed as pending a terms review rather than promised.
- OCR for scanned PDFs — a
no text layerrefusal is the honest answer for now.