Vinci by SimpleDirect®
AI with character — that you own.
Vinci is an open-weight AI with a trained personality you'll enjoy using — grounded in your own documents, backed by a public Constitution, and hosted in Canada with Zero Data Retention. Vinci 1.0 (vinci-piccolo) launches June 30, 2026.
- Public Constitution…
- Public Character doc…
- Open weights · yours to run…
- Adversarial test results…
Early access is waitlist-only until launch.
Vinci vs the default
A different kind of model.
A closed box you take on faith.
A public Constitution, Character, and adversarial results — with open weights, so you check it yourself.
Confident answers built on invented citations.
Answers tied to your own documents — and it refuses to invent a source.
An API that can be revoked, repriced, or throttled.
Open weights — self-host it, on Canadian sovereign infrastructure, Zero Data Retention by default.
What Vinci is
Five properties.
Not a slogan — a specification. Each one is something you can check.
Capable
Built on frontier-competitive open-weight foundation models.
Character
Trained against a public Constitution and Character document.
Verifiable
Public adversarial test results. Run your own against the open weights.
Grounded
Answers tied to your own documents through retrieval, not fluent guessing.
Sovereign
Hosted on Canadian infrastructure with Zero Data Retention by default.
Grounded, not guessing
It cites your documents — or says it can't.
Most models fabricate a citation rather than admit a gap. Vinci grounds in what's actually retrieved, and refuses to invent.
Most AI
Vinci
The synthesis principle
What Leonardo could be in 2026.
Vinci is what Leonardo da Vinci could be in 2026 — combined with the substrate-operator values SimpleDirect built from years of watching what AI can and can't do for people.
Leonardo's discipline of careful observation — saper vedere, "knowing how to see" — is the foundation: structured habits, sustained curiosity, recursive notebooks, and real humility about his own limits.
That is what makes Vinci good. It grounds answers in what is actually in your documents before it pronounces — instead of generating fluently from training, without verification.
His second foundation was ethical: how the powerful treat the less powerful. Operators hold power over a customer's data and workflow. We treat that with care, not extraction.
So the Constitution is public. The adversarial test results are public. The weights are open. You don't trust the claims because we ask — you verify them, because we make that possible.
The family
Five models, one methodology.
The family follows Leonardo's creative process. Vinci 1.0 is vinci-piccolo; the rest follow through 2026.
Open, permissively-licensed base + Constitutional Fine-Tuning
Open, permissively-licensed base + Constitutional Fine-Tuning
Open, permissively-licensed base + Constitutional Fine-Tuning
Open, permissively-licensed base + Constitutional Fine-Tuning
Open, permissively-licensed base + Constitutional Fine-Tuning
Yours to own. Yours to verify.
The Constitution, the Character document, the methodology, and the adversarial test results ship in the open. If Vinci's behaviour ever diverges from them, that's a bug — and we treat it like one.
Launching June 30, 2026
Follow the build.
Open weights, a public Constitution, and a model you can verify — the day it ships. Get the launch in your inbox. No spam, just the drop.
or follow along — Follow on X