Our productLaunching soon
Spitometro
An AI analyst that assembles an evidence-cited dossier on a specific home before you buy, sell or rent.
Service II of 5
Assistants and agents that answer, plan and act inside your product, built with evals and guardrails.
Most of the value AI brings to a business comes from narrow, well-integrated systems: an assistant that knows your catalogue, an analyst that assembles a report, a workflow that reads documents so people don't have to. We build those, and we build them so the language model writes the language while numbers, prices and decisions come from code you can test.
Good fit if you are
01
What you get, phase by phase.
02
Six rules every assistant and agent we ship follows — including the one on this site.
Start with the simplest structure that works, and add autonomy only where it pays for itself.
Prices, scores and facts come from deterministic modules you can test. The model explains them.
A fixed set of real questions runs before each deploy; regressions block the release.
Scope limits, refusal behaviour, rate and cost caps, and a clear path to a person.
People are told they're talking to an AI (EU AI Act, Article 50), and model providers and data flows are documented.
Claude, Gemini or open models, chosen per task and kept swappable.
03
Where this service shows up in projects we've shipped.
04
The ones we hear most about this service.
Mostly Anthropic's Claude and Google's Gemini, chosen per task for quality, speed and cost. Systems are built so the model can be swapped without rewriting the product.
It can, which is why we narrow its scope, ground it in your own content and data, keep numbers out of its hands, and test it against a set of real questions before every release. When it doesn't know, it says so and offers a person.
Chatbots must tell people they are talking to an AI (Article 50, applicable since 2 August 2026). Our assistants do that by default, and we document what the system does, which providers process data, and where.
It depends on volume and model. We design for cost from the start — smaller models where they're good enough, caching, hard daily caps — and show you the running cost per conversation before launch.
Other services
Ask about this
AI answer from Neumatics' assistant. It only knows this site, can make mistakes, and nothing is stored.
Next step
Tell us about it in a few short steps, then pick a time for a 30-minute call. An idea, a problem or an old site is enough to start.