Upload docs & ask

This walkthrough takes you from an empty workspace to a grounded, cited answer over your own documents. It assumes you've added a model provider (an LLM and embeddings).

1. Create a knowledge base

A knowledge base is a shareable collection of documents Fosnie can retrieve from. At the top level it's called a Library; inside a Project it's Project Knowledge – the same thing.

Create one from the Libraries area (or inside a Project), give it a name, and open it.

2. Add documents

Upload files by drag-and-drop. Supported inputs include PDF (including scanned), DOCX, XLSX, PPTX, images, and plain text. The default size limit is 100 MB per file.

Ingestion runs in the background – the upload returns immediately and each document shows a live status: uploading → extracting → indexing → ready (or error, with a reason). Under the hood Fosnie extracts text (OCR for scans and images), splits it into overlapping chunks, and embeds it into the vector store. You can start asking as soon as documents reach ready.

3. Ask

Open a chat with the knowledge base attached and ask a question in plain language. Before the first token you'll see a short "searching your library" activity while Fosnie retrieves.

Behind that, an agentic retrieval loop runs entirely on your server:

1. Decompose your question into sub-questions.
2. Hybrid search (dense + keyword) across the knowledge base.
3. Rerank to the most relevant passages.
4. Grade whether the passages actually cover each sub-question – and search again if not.
5. Answer once, over the assembled evidence.

4. Check the sources

The answer streams with inline citations. Click one to open the exact source document, jump to the passage, and see it highlighted – so every claim is traceable back to your material.

Next steps

• Deep Research – multi-step, fully-cited investigation over your files or the web.
• Document work – generate and edit DOCX/PDF/XLSX with tracked changes.
• Connectors & MCP – pull documents in from Gmail, Outlook, or your DMS.