Wind
Hurricane · tornado · convective stormHurricane is weighed by its PML at the 1-in-100 and 1-in-250 return periods. Tornado and convective storm are secondary perils that build up in the aggregate.
The future, with reason.
An exclusive, closed, private AI, built by daryxlabs in conjunction with InsurEco and other leaders in the insurance industry.
Tap a tier to read about it, or replay the question.
What reasons and acts. It plans a piece of work, shows you the plan, and carries it out inside the fence you set.
What it knows about you. JCI holds your organization's knowledge graph and hands Chaac only what a question needs.
What runs beneath. Frontier and local models, routed by difficulty: small questions to small models, hard ones to large.
Chaac draws on PolicyBlockchain, a decentralized insurance data chain that brings loss, weather, form, referral and market data together in one secure chain. The data is already there. The insurance smarts are already dialed in.
Picture a car that drives better because it has learned from millions of miles on real roads, rather than from reading the owner's manual.
A general-purpose AI has read the manual. Chaac is grounded in the road: the losses, the storms, the forms, the referrals and the market, all in one secure chain, with the insurance smarts to read them.
Built to be more accurate and more efficient on insurance work than a general-purpose AI.That is the design intent. The chain below is what it stands on.
Events on chain
175,079
hash-linked events since April 2026
Market
595,528
quote indications
Submissions
121,202
risks submitted
Policies
100,748
policy records
Insureds
62,520
insured entities
Agencies
7,596
agencies in the registry
Carriers
1,618
carriers and companies · 245 programs
An AI you can see into, working inside walls you drew. Those two things together are what make it fit for insurance.
Before Chaac acts, it shows you the plan: the steps it intends to take, the questions it has for you, and what the work will cost.
Chaac works within boundaries your organization sets: which data it may read, which actions it may take, and whose records it may touch.
Chaac is the Maya god of rain and lightning. Ours is an AI, and more modest: it cannot call off a storm. But it speaks the language of the people who price, place and carry the risk.
He is the rain god. Of course he knows the perils.
Hurricane is weighed by its PML at the 1-in-100 and 1-in-250 return periods. Tornado and convective storm are secondary perils that build up in the aggregate.
Surge travels with a named storm, but most property forms treat it as flood: excluded, sublimited, or written on its own.
Rarely the 1-in-100 headline, often the steady drain on the annual aggregate. Roof age and the wind and hail deductible decide much of the loss.
He carries the lightning axe, so this one is personal. Wildfire accumulations are managed zone by zone, and parametric triggers are one way to place them.
Frozen pipes and roof collapse arrive together across a region, which is why occurrence definitions and hours clauses matter.
Modelled and actual. The curve, the tail, and how the tail gets paid for.
Who owns the risk, and how it is fronted, ceded, cell by cell, and reported.
Where the limit comes from, what it costs, and when the market turns.
A lexicon, not a feature list: the terms Chaac reasons in, carved for the people who already use them.
Each one alone is useful. Together they make something no general-purpose AI has.
A model is what reasons. A context layer is what it reasons about. They are different things, and keeping them apart is the point.
Jefferson holds your organization's knowledge as a graph: insureds, policies, carriers, programs, claims, and how they connect. When a question arrives, it assembles only the context that question needs and hands Chaac that, rather than the whole file room.
The model reads less, so it costs less, and simpler questions can be answered by smaller, cheaper models. And your organization's knowledge stays in its own context layer, not inside anyone's model.
What the model reads · illustrative lengths, not measured figures
Two capabilities are being carved now. Neither is available in this release.
Coming · not yet available
The base, made yours
Choose which models run beneath Chaac, frontier or local, for your organization or for a single kind of work. Jefferson's context stays the same whichever model reads it.
Coming · not yet available
The temple, on your ground
Run Chaac, its context layer and its models inside your own environment, so the whole temple stands within your walls.
We build your organization a custom MCP on your own data. Then you reach Chaac and Jefferson through whichever door suits the work.
Door I
The workspace. Chaac and Jefferson built into the place your team already works.
Door II
From Claude, Cursor or any MCP client. Your data and your context, in the tools you already use.
Door III
A chatbot, for the people who simply want to ask.
Chaac is being opened to a small number of organizations at a time. Begin with an assessment of where AI fits your work, or write to us directly.