RAG Knowledge Base

March 23, 2026 · View on GitHub

Upload documents so JARVIS can cite them when answering questions.

Prerequisites:

  • JARVIS running, logged in, and an OpenAI API key configured in Settings (required for embeddings via text-embedding-3-small)

Upload a Document

  1. Click Documents in the left sidebar.
  2. Click Upload Document.
  3. Select a file — supported formats: PDF, TXT, MD, DOCX.
  4. Wait for processing. JARVIS chunks the document (500 words / 50-word overlap) and indexes it in Qdrant using text-embedding-3-small (1536 dimensions).

Ingest from a URL

  1. In the Documents panel, click Add from URL.
  2. Paste a public URL (HTML page, PDF link, etc.).
  3. JARVIS fetches, extracts, chunks, and indexes the content automatically.

Ask a Question

After uploading, start a new conversation and ask a question related to your documents:

"Summarize the key findings in my Q4 report."

JARVIS performs a hybrid vector + keyword search, retrieves the top-5 matching chunks, and injects them as context before the LLM call. Sources are cited inline: [1] "document-name".

Workspace Collections

Within a Workspace, all members share a common knowledge base:

  1. Go to Workspace Settings → Documents.
  2. Upload documents — they are stored in a shared workspace_{id} Qdrant collection.
  3. All members' conversations in that workspace automatically search both their personal collection and the workspace collection.

Personal and workspace results are merged and re-ranked by combined score (70% vector, 30% keyword overlap) before being passed to the LLM.

Remove a Document

  1. Go to Documents.
  2. Click the trash icon next to the document.
  3. The document and all its indexed chunks are deleted.