How Ounie Works
Ounie combines two ideas usually pitched as alternatives: a self-building wiki and retrieval at scale. The result is grounded, cited answers from your own material.
Three steps
- Add anything. Upload files, paste links, jot notes, or add audio and video. Ounie processes each source in the background — OCR for images, transcription for audio and video, content extraction for web pages and documents.
- It builds your wiki. An llm reads each source and writes a clean, cross-referenced wiki page. Key entities and concepts get their own pages, and everything connects through
<a class="wikilink" data-slug="wikilinks" href="/b/ounie/how-ounie-works/wikilinks">wikilinks</a>into a living knowledge graph. You don't organize anything manually — the wiki maintains itself. - Ask anything. Open the Ask console for a brain and type a question in plain language. Ounie searches your brain, gathers the most relevant pages, and answers using only that context — citing every claim.
The architecture, in plain terms
Ounie uses a hybrid llm-wiki + rag design:
- LLM-Wiki: every source you add is read by an LLM and synthesized into a coherent wiki page, so knowledge compounds instead of being re-discovered on every query.
- RAG at scale: those wiki pages are chunked and embedded into a vector index (pgvector), so finding the right pages stays fast even with thousands of pages per brain and across millions of users.
What happens when you ask
When you ask a question, Ounie vector-searches your brain, pulls the full wiki pages for the best matches plus their linked neighbor pages, and answers from only that context. Then it cites every claim back to the page it came from. If the relevant material isn't in your brain, Ounie says so rather than inventing an answer.
Why the hybrid approach beats "chat with your docs"
Most "chat with your docs" tools just chunk, embed, retrieve, and hope. Ounie keeps a coherent, cross-referenced wiki and adds vector search — giving you the trustworthiness of structured pages, the scale of retrieval, and a knowledge graph for free. Because the wiki is built once and reused, answers get faster and richer as the brain grows rather than slower.