A knowledge base that cites its sources — and says when it doesn’t know.
Most chatbots answer everything, confidently, whether or not they have the facts. This one only answers from a fixed set of documents, shows you the exact passage behind every claim, and refuses when the answer isn’t there. Ask it something below.
Why this matters for your business
The question is never “can AI answer this?” It always can. The question is whether you can trust the answer enough to put it in front of a customer, or to act on it.
Every claim has a receipt
Each answer carries the source passage it came from. When a colleague asks “where does that number come from?”, there is an answer, and it takes one click.
Refusal is a feature
Ask it about the weather and it will tell you the answer is not in the knowledge base. A system that admits the gap is worth far more than one that invents something plausible.
The quality is measured
Not claimed — measured, on a fixed question set, and published. That is how you tell an engineered system from a demo that happened to work the day someone recorded it.
Why it’s called Mimir
Mímir is the figure in Norse myth who guards the well of wisdom. Odin gave an eye for a single drink from it — a fair price for answers you can trust. The name is the whole design brief: this system draws from one well and will not pretend to know what is not in it. The mark above is the rune, with the well beneath it.
Everything here is open
The code, the document set, the question suite and the results are all public. You can read how it works, run it on your own documents, or check the numbers yourself — including the model comparison behind the choice of which model runs in production, and what each one costs per query.
Want this on your own documents?
Contracts, product manuals, internal policies, support history — the pipeline is the same. The work is choosing what belongs in the corpus and deciding what the system is allowed to say.
Tell me what you would ask it