AI & agents · working demo

Intake agent

Enquiries arrive as messy paragraphs. This reads one, pulls out the structure, scores urgency, drafts a reply, and stops to ask you when it isn't confident. Paste anything, or try a sample.

Incoming enquiry

Pipeline

Extracted

Nothing processed yet.

What's real here: the extraction genuinely runs, entity patterns for phones, emails and addresses, keyword-weighted urgency scoring, service-area matching, and confidence thresholds that decide whether the agent proceeds or escalates. Type a vague message and watch it correctly refuse to guess. That refusal is the most important behaviour in the whole demo.

What's not: this runs a deterministic rules engine in your browser, not a language model, so it's fast, free, and shows the logic honestly. In a real build the fuzzy extraction is a model call and everything else stays exactly as ordinary code, which is the point I make on the AI page: use a model for the genuinely fuzzy part, and code for everything with a correct answer.

Drowning in enquiries?

The win isn't replacing you, it's that every message is read, sorted and drafted the moment it lands, and you only touch the ones that need judgement.