Built different. Verified by design.
Compass & Cove builds Cove — a grounded, verified, low-hallucination intelligence that runs entirely on our own hardware, with no borrowed model weights.
A desk in Minnesota. Not a datacenter.
Compass & Cove was founded by Mark Brush in Minnesota. No venture capital, no datacenter, no team of hundreds — one builder, consumer hardware, and a thesis: trust comes from verification, not from scale.
Cove runs on machines we own and operate ourselves. Your questions are answered on our hardware — not relayed to a third-party AI provider. There are no borrowed model weights anywhere in the answering path.
weights: our own — no borrowed or licensed third-party models
answering rule: cite, verify, or say so honestly
third-party AI APIs in the answer path: none
Verification first. Words second.
Most AI predicts the most plausible-sounding next words. Cove works the other way around: a question is decomposed into checkable claims, each claim is verified — cross-checked across independent sources, actually computed when it's math, proven step-by-step when it's logic — and only then is the answer written. With receipts.
Sources, not vibes
Factual answers are cross-checked against independent sources before they're served. Agreement across sources raises confidence; disagreement gets surfaced, not hidden.
Math is computed, logic is proven
Calculations are actually executed and logical chains are derived step by step — not pattern-matched from text that merely looks like reasoning.
Gaps are stated, not papered over
When verification fails, Cove says "I can't verify that, so I won't guess." Low-hallucination is an engineering target here, enforced by gates — not a slogan.
Growth is tested before it ships
Cove's knowledge and skills grow continuously — and every improvement has to pass verification gates before it reaches customers. Failures roll back automatically.
The rules she answers by.
- 01 Cite, verify, or say so. Every answer is grounded in checkable sources, computed, or proven — or Cove tells you plainly that she can't verify it.
- 02 No deception. Cove never presents fiction as fact, inflates her confidence, or papers over a gap with plausible-sounding filler.
- 03 Your conversations are yours. We don't sell your data and we don't use your private conversations to train models. Processing happens on hardware we own.
- 04 No harmful output. Cove operates under a hard safety constitution that can't be talked around — no scams, no exploitation, no harm.
- 05 Receipts on request. Answers carry verification labels — what was checked, against how many sources, with what confidence — so you can audit instead of just trusting.
- 06 Honest marketing too. We say "low-hallucination" and mean it as an engineering discipline. We will never claim "never hallucinates" — nobody can honestly claim that.
The best way to understand Cove is to test her.
Ask her something you think she'll get wrong. Three questions a day, free, no signup.
Try Cove free