Your first chat
1. Get documents in
Documents → Upload, or drag files onto the chat box to attach them to the conversation you're in.
Supported: PDF, Word, Excel, PowerPoint, plain text and images. Scanned PDFs and photos of pages are read with OCR, so they end up as searchable as anything else — they just take longer to process.
A document moves through Processing to Ready. Only Ready documents can be used in an answer. Large scanned files are the slow case; on a self-hosted CPU-only server they can take several minutes.
2. Ask a question
Open Chat and type. The first answer after an idle period is the slowest one on a self-hosted server — the model has to load into memory first. Subsequent questions are quicker. If that first wait is painful, see Performance and memory.
3. Scope the conversation
Above the prompt box you can narrow what the answer is allowed to draw on:
- All collections — everything you have access to.
- A collection — one project or client.
- Specific documents — pick them individually.
Narrower scope means faster, sharper answers and far less chance of the model pulling in something irrelevant. If you know the answer is in three documents, select those three.
4. Read the citations
Every answer carries citations back to the source. Open one and you land on the document and passage the sentence came from.
5. Come back to it
Conversations are saved under History, searchable and filterable by date and status. Start a New chat when the subject changes rather than continuing a long unrelated thread — a focused conversation gives better answers.