Unified Patent Court · Case law research
UPC Agent is a research tool for patent litigators. It reads the judgments and orders of the Unified Patent Court and answers questions about them in continuous prose, with every proposition tied to the decision it came from. It is built for the moment when you need to cite a court, not skim a headnote.
Access is invitation-only during the closed beta.
Keyword search returns the decisions in which a term appears. That is a different thing from an answer. These are the kinds of questions the tool is designed to take:
On what reasoning have Local Divisions granted provisional measures without hearing the defendant?
Compares the ex parte orders across divisions, sets out the urgency and proportionality tests each applied, and marks where the Court of Appeal has since narrowed them.
How does the Court of Appeal weigh the balance of interests in applications for preliminary injunctions?
Traces the factors the Court has actually named — likelihood of invalidity, delay by the applicant, irreparable harm — with the passage each one comes from.
What standard do the divisions apply to added matter in revocation actions?
Sets the Central Division's approach against the divergences in the Local Divisions, and shows where the reasoning follows established EPO practice and where it departs from it.
How has a FRAND defence been treated when raised alongside an infringement claim?
Follows the defence through the UPC decisions and the German national judgments that shape it, including the negotiation-conduct requirements each court imposed.
Retrieval runs over the text of the decisions themselves — the reasoning, the orders, the procedural history — rather than over headnotes or metadata. A question about the treatment of a defence therefore reaches the paragraphs where the court discussed the defence, whether or not the parties' chosen terminology appears in them.
Every factual sentence in an answer carries a numbered footnote to the decision it rests on. The footnote is the only place a case identifier appears: docket numbers are never written into the prose, because a case number invented mid-sentence is the single most damaging thing a research tool can produce. Each answer is checked against that rule before it reaches you, and an answer that cannot support a proposition from the corpus says so instead of reaching for a plausible substitute.
A language model performs the synthesis. Retrieval and ranking sit outside it, and every answer passes a rule-based citation check before it reaches you.
The corpus is drawn from the published case law of the Unified Patent Court: decisions and orders of the Court of Appeal, of the Local and Regional Divisions, and of the Central Division. It covers infringement actions, revocation actions and counterclaims, applications for provisional measures, and the procedural orders that shape them under the Agreement and the Rules of Procedure.
German national patent case law sits alongside it, so that a question which turns on doctrine developed before the Federal Court of Justice and the Higher Regional Courts can be answered across both. Answers say which court and which instance each proposition comes from.
Patent litigators, in-house counsel running European patent portfolios, and the associates who have to have read everything by Monday. It is a research aid: its output is informational and does not replace independent legal analysis, and every decision it cites should be verified against the official source before you rely on it.
Access is currently invitation-only. Sign in with the work or personal account we have on the allowlist.