What is a legal ontology?
A legal ontology is a formal structure for a case or corpus — the entities (parties, issues, claims, statutes, facts) and the relationships between them, written down precisely enough that software and people can reason over the same model. "Context" is an informal way to hand an AI information; an ontology is a formal one. The distinction is the whole point of doing legal automation rigorously rather than just fluently.
Why does "formal" matter?
Today's language models learned language the way humans evolved to use it — by intuition, not by rule. They are brilliant at sounding right, and were never built to be rigorous. But civilization doesn't run on pure intuition; it runs on formal systems. Mathematics is the universal one; law and engineering are specialized ones. A formal structure is how you get from "sounds plausible" to "can be checked."
Does a formal system hand you the answer?
No — and this is what the "ontology" buzzword usually hides. Even mathematics is incomplete: there are true things no formal system can prove. Law is something else again — it is interpretable; it lives in argument. So formalizing a case isn't about reaching a final, settled answer. It's the work of structuring a question precisely enough that experts can reason about it — and disagree about it — clearly.
Whose job is the ontology — the lawyer's or the agent's?
The structuring and interpretation belong to professionals. The agent's job is to help build the ontology faster — by onboarding clients and data, proposing structure, and keeping it consistent — not to replace the judgment that makes a case sound. The future of legal automation isn't a smarter machine that decides; it's a smarter system where the people who interpret the law can execute, collaborate, and reason faster.
How does Relex use an ontology?
Relex gives professionals a shared formal structure for each case and corpus. It maps the matter as a hub around a single matter node — parties, the legal issues and claims at stake, the statutes those claims cite, the confirmed facts, and the open questions — and the verbs that bind them (modifies, triggered by, governs, supports, contradicts). It is built automatically over anonymized data only, so the agent reasons over grounded facts instead of free-associating, and you and your colleagues argue over a model you can all see. In the Partner Program, professionals get a new market of clients to do exactly this work.
Try it now — with Claude or standalone
Build the formal structure of your next matter faster, without exposing client identities.