Chat with your documents, kept in Australia

Chat with your documents in AusGPT by uploading a PDF, Word or image file and asking questions about it in plain English. Claude reads the document and answers from it: summaries, comparisons, extracted details, first drafts. It's the fastest form of AI document analysis for everyday work, and your files are stored and processed in Australia.

How document chat works in AusGPT#

1. Upload your file. Add a PDF, Word document or image to a conversation. Contracts, reports, policies, letters, forms, screenshots and photos of printed pages all work as starting points.
2. Ask a question. "What are the termination clauses?" "Summarise this report in five dot points for the board." "List every date and deadline."
3. Claude answers from the document. The file becomes part of the conversation, so Claude can quote it, compare it with another upload, or use it to draft something new.
4. Check against the source. Ask Claude to quote the passage it relied on, then confirm it in the original.
5. Save prompts that work. Keep your best document prompts in the shared prompt library so your team gets consistent results.

"Train ChatGPT on your own data" usually means something simpler#

People often search for ways to train ChatGPT on their own data. In most cases what they want isn't training at all. It's giving the AI their documents to read, so it answers from those documents rather than from general knowledge.

AusGPT never uses your documents to train AI models. Each conversation reads the files you give it, and Amazon Bedrock doesn't share your prompts or documents with Anthropic or any other model provider.

What you can do when you chat with your documents#

• Summarise. Turn a 60-page report into a one-page brief, or a long email chain into decisions and actions.
• Compare. Upload two versions of a contract or policy and ask what changed, or compare two supplier quotes line by line.
• Extract. Pull dates, amounts, parties, obligations or requirements into a table you can paste into a spreadsheet.
• Draft. Write a reply letter, a client summary, an FAQ or a training handout based on the document.
• Check. Ask what's missing from a tender response, or where a draft policy contradicts the procedure it replaces.

Use cases by industry#

Prompts to try#

Copy any of these into AusGPT after uploading a file. They work in any AI assistant that accepts documents.

Read the attached document. Summarise it in [number] dot points for [audience]. Then list any deadlines, amounts and named obligations in a table with the page or section where each appears. If something isn't stated in the document, say so rather than guessing.

Compare the two attached documents. List every material difference in a table with columns: topic, version A, version B, why it matters. Quote the exact wording for each difference.

From the attached [document type], extract [the fields you need, e.g. party names, start date, fees, notice periods] into a table. Leave a cell blank and write "not found" if the document doesn't contain it.

Using only the attached document, draft a [letter/email/summary] to [recipient] explaining [topic] in plain English. Keep it under [number] words and flag anything you weren't sure about.

Privacy: where your documents go#

• Stored in Australia. Uploaded files and conversations are stored on Amazon Web Services in Sydney.
• Processed in Australia. When Claude reads a document, the request goes through Amazon Bedrock's Australian cross-region inference profile, which routes requests only between Sydney and Melbourne.
• Not used for training. Your documents aren't used to train AI models, and model providers have no access to your prompts and documents on Bedrock.
• Shared when you choose. You decide which conversations to share with colleagues in your team workspace.

Location is only part of the picture. Only upload documents you're allowed to use this way. If a file contains personal information, the Privacy Act still governs how you use it, and some material, such as documents covered by court orders or confidentiality obligations, may have its own restrictions. Our guide to secure AI for business covers what to check.

Honest limits#

Document chat is fast and useful, but it isn't infallible:

• Check answers against the source. AI can misread a clause, miss a qualification or state something the document doesn't say. Anthropic's own guidance on reducing errors recommends asking for direct quotes and letting the model say "I don't know". The prompts above do both.
• Image quality matters. Blurry scans, handwriting and complex tables may be read imperfectly. Check extracted figures before you rely on them.
• Ask focused questions about long files. With long or multiple documents, ask about specific sections or topics rather than "tell me everything".
• It's not your records system. Keep the original documents in your usual system. AusGPT is where you work with them, not where you file them.

See everything else AusGPT does on the features page.