Search UK case law down to the paragraph — on your own hardware.
Trialist indexes every judgment published by the National Archives' Find Case Law service and returns the individual numbered paragraph you need, not a pile of documents. Behind it sits a measured research programme on procedural law and citation integrity. The whole stack runs inside your chambers or firm, so no client query ever leaves the building.
What it does
Pinpoint results
Hits are numbered paragraphs — the unit you actually cite. One click opens that paragraph inside the full judgment, and copy cite gives you a paste-ready pinpoint reference.
Search by proposition
Describe the point of law in plain English. Semantic search finds the passage even when it shares no keywords with your query, while case names, NCNs and "quoted phrases" still match exactly first.
Every citation is a link
Citations inside a judgment open the cited case where we hold it, and link out to Find Case Law where we don't. Your research trail is kept in your browser only.
Runs on-premise
Index, embeddings and interface all run on your own machine or a box in your server room. Nothing is sent to a third-party cloud — which is what makes confidentiality duties easy to answer.
What a result looks like
Illustrative layout only — live results come from the published corpus.
Built on measured research, not assertion
Search is the part of Trialist you can see. Behind it sits a research programme on the pre-court procedural law that decides cases before the merits are ever reached — limitation, service, pre-action conduct, disclosure, relief from sanctions, abuse of process.
A procedural corpus
52,000 judgments held; ~10,000 structured procedural findings extracted per production run, every one gated by schema validation, verbatim anchor quotes and paragraph-existence checks.
Honest evaluation
A frozen holdout of 3,112 cases, split by date rather than at random, so we measure whether the system generalises to law it has never seen — not whether it memorised.
Fabrication, measured
We built a citation benchmark instead of asserting safety. On identical questions a bare 32B model invented ten authorities and fell for every trap; Trialist invented none and refused all five.
Why this matters more than it sounds
A larger model can avoid inventing citations simply by declining to cite anything — and one we tested did exactly that, offering one of fifteen expected authorities. Trialist returned all fifteen, with zero fabrications. Being useful and being reliable are separate problems, and both have to be solved.
Licensed, and deliberately narrow
Trialist is operated by Executive Force AI Ltd as a Find Case Law transactional licensee, and is built to the exact purpose of that licence: to publish the current version of judgments and provide search facilities to find and access them. Takedowns and amendments are enforced when a page is served, not just when the corpus is built, and Crown copyright is attributed throughout.
What Trialist does not do
- No outcome prediction and no judge profiling.
- No generated legal advice or AI-written answers.
- No citator "overruled / applied" verdicts.
Everything on screen is a passage from a published judgment that you can read and check for yourself.
See it on your own research
Demos are run against the live corpus and take about twenty minutes — bring something you looked up this week. We are working with barristers' chambers, litigation firms and costs lawyers on early deployments.