The same engine reads the working documents, watches the hearings, and reads the press review. You read one answer, with the source attached. That's the whole product.
Ask about a procedure, a rapporteur, a vote, an amendment, a hearing. Get the answer with the document — or the moment in the video — sitting next to it. If we don't have a source, we don't have an answer.
An answer is only as good as what it's standing on. Thembi reads the working documentation of EU law-making from the earliest preparatory step through delegated acts — the phases where positions are still moving and strategy actually matters. That's the substrate every Thembi answer is grounded in.
We parse messy, inconsistent EU APIs and extract the substantive content from unstructured text — so the document you see is the document you can ask questions of.
Council discussions, committee opinions, delegated acts: the same event gets tagged five different ways across institutions. We disambiguate, so the timeline you read is the one that actually happened.
Who voted with whom, which group changed its mind between committee and plenary, which Member State suddenly stopped abstaining. The pattern behind the vote, not just the result.
A built-in press review, but read by an engine instead of by you. We read across outlets, strip opinion from reporting, and keep what's verifiable. So when "Brussels sources say…" turns up in the morning, you already know whether it's a position, a leak, or a guess.
EU committees, plenaries, midday briefings, press conferences — hours and hours of livestream a week. Thembi watches all of it, so when your topic comes up you get the clip and the transcript. You read the moment, not the meeting.
High-accuracy speech-to-text across the EU's working languages, with speakers identified.
Every segment summarised by policy relevance — including the obscure press conferences nobody else watches.
"Did anyone object to the Commission's position?" "What did the rapporteur say about timing?" Ask in plain English, get a sourced answer with a timestamp.
A Virtual User Persona is a perspective on the same underlying engine. It decides what reaches you, what gets summarised, and what gets ignored — based on the role, sector or client it's trained on. Consultancies run one per client. In-house teams sharpen one over time. Same answer engine, twelve different signals out.
Our prompt scaffolding is shaped by EU lawyers, not generalists — so when a persona says "competition file" it means competition file, not anything that includes the word.
Documents, livestreams, press, votes — read continuously, kept with the original source attached.
Every event tagged, every procedure assembled into a timeline you can actually read.
Your persona keeps what's relevant to your role, sector or client — and drops the rest. Less noise, not more dashboards.
Ask a question. Get a sourced answer. Or get a briefing in your inbox before you've asked.