Termus was an AI search engine for Dutch case law, at search.termus.ai. The pitch was simple: faster, smarter search over rechtspraak, with AI on top that answered the question instead of handing you twenty links to read. There was a second app too, app.termus.ai, for legal Q&A, upload a document and ask about the case law behind it. I ran the whole thing through my company Selitic in 2024: the domain, the OpenAI bills, the Azure credits, a data processing agreement with [customer] for the data. I did the tech, mostly part time. [...] did sales and walked into law firms. [...] did business and the deck.
Why it didn't work
The problem was never the demo. The problem was that everyone had the same demo. I kept a list of competitors and it only grew: Harvey, Leya, Legalfly, Qura, and a dozen more, plus whatever launched that week. At some point it felt like we were the seventeenth legal-chat company. The category was full of funded teams all shipping the same impressive first ninety seconds.
We got to a working product fast. Call it 20% of the work to get 80% of the way there. But the last 20%, the part where a lawyer trusts the answer and the chat flows hold up for real questions, is most of the actual work, and it is the part that matters. A good legal-AI demo is cheap. Something a professional will rely on is not.
Underneath that was a structure problem. I was part time and felt responsible for everything at once: tech, customers, business, the subsidy paperwork, the product. The people putting time into sales were not seeing the customers or the funding show up to match it, which was fair. And I was stuck in a familiar loop. I would only go full time with funding, and funding wanted a full-time founder. Chicken and egg.
Calling it
By autumn we were openly asking whether to pivot. Dial the chat back, focus purely on making search excellent, strip it to one app that does upload plus search, maybe license the search itself as an API. Google for Dutch jurisprudence. We set a GO/NOGO for December and stopped chasing new customers until there was funding or a second person going full time.
There never was. It did not find a market and I wound it down. Not much survives, mostly the Dutch datasheet. What I kept is the lesson: in a crowded category the demo is the easy part, and running part time against funded full-time teams is a losing structure no matter how good the tech gets. I do not regret building it. I learned exactly what I would want in place before starting the next one.