A frontier model is a component. HAQQ is the system around it.

A general assistant optimises for one thing: what is this person asking, and what is a good reply. That is the right goal for almost every other use of AI.

Legal work needs a different opening move. Which legal issue is actually in play, which jurisdiction governs it, what language is that law written in, which sources bear on it, and how does the answer need to be structured to be usable in a matter. HAQQ resolves those before a model generates anything, which is what we mean when we say it approaches a request the way a lawyer does.

People do not use a legal assistant the way they use a general one. They upload the dispute, the client file, the internal memo, the strategy. The sensitivity of that material is a property of the work, not a setting someone forgets to switch on.

So encryption, access control, isolation between matters, anonymisation before processing and data residency are how the system is put together rather than features bolted around it, and the HAQQ terms governing a deployment exclude customer data from model training.

HAQQ routes different parts of a legal task to different models based on what each is good at. A frontier model still performs the final generation, and we say so on the engine page rather than implying we trained our own.

What we own is the part that carries the legal value: orchestration, retrieval, jurisdiction logic, memory, guardrails and grounding. Those run the same whichever model sits underneath, so adopting a better or cheaper one is a configuration change rather than a rebuild.

That is also why frontier progress helps rather than threatens. A stronger model makes HAQQ stronger without making it redundant, because the layer that understands law is not the layer being replaced.

HAQQ is used by practising lawyers and by people who have never spoken to one, and the same legal question deserves a different answer for each.

A lawyer wants authorities, reasoning they can check, risk analysis and language they can draft from. Someone reading their own lease wants to know what it means for them and what to worry about, in words they already use. The analysis underneath stays the same; the register changes.

A general assistant cannot make that adjustment reliably, because nothing in the product tells it which of the two it is talking to.