FAQ
Straight answers.
What do you actually build?
Three products, and custom systems for clients to the same standard. Front Desk handles conversations: an AI receptionist (voice and booking), a support bot that answers from your documents, and lead qualification that scores and routes enquiries, as modules on one engine. Docket handles documents: invoices, receipts, purchase orders and credit notes turned into checked data, with a risk gate against bank-detail fraud, duplicates and purchase-order mismatches. Dossier checks applicants' documents (bank statements, bills, payslips, IDs) for lenders, letting agents and firms with AML duties, and catches edited or borrowed ones. If you need something none of them does, we build it.
Can you build something custom for us?
Yes. Everything on this site is our own work running in production, and we build custom AI systems the same way: assistants on your data, document pipelines, voice and chat agents, decision engines, automations and integrations, or a rescue of a bot that misbehaves. A scoping call, a written scope and a fixed quote come first; weekly demos, tests and a written handover follow.
Can you process our invoices?
Yes, that's Docket. Every value is located on the page it came from and checked by code (line items, totals, VAT, IBAN, dates), a changed bank account or a look-alike supplier stops the document, and verified data goes to Excel, CSV, a webhook, Google Sheets or your accounting system. You can try it with your own invoice on the Docket page.
Can you spot a forged bank statement?
That's Dossier. It examines the PDF itself (a version saved after issue, an amount retyped over a white box, a different font, a statement remade in Word), adds up every balance, checks bank details and a passport's MRZ, and compares names, addresses and dates with what the applicant declared. It says what is wrong with a document, with the place on the page; it never scores the person. You can forge a synthetic applicant yourself on the Dossier page.
Do I need a new website?
No. The system plugs into what you already have: a script tag for chat, your existing contact form, a forwarding rule for your email, a Telegram bot, a voice widget and your phone number.
Does it work with my CRM?
Yes, through one signed webhook: every lead, booking, form and call is sent as JSON to a URL you choose, so HubSpot, Pipedrive, Zapier, Make, n8n or a Google Sheet can take it from there. Ready n8n recipes for HubSpot, HighLevel (calendar included) and Google Sheets check the signature first and are tested against real accounts. Your team can also work straight from the built-in inbox.
What happens when a customer wants a real person?
The assistant tells them the team will join, you get an alert, and you reply from the inbox: by email, or live into their website chat or Telegram. You hand the conversation back to the assistant when you're done.
How do you stop the AI from making things up?
Hard facts such as hours, prices, services and contact details come from a config file, not from the model. Support answers must be supported by the retrieved documents; an answer containing a price, time, phone number or email that isn't in them is blocked. When the documents don't cover a question, the bot says so.
Can someone trick it with a prompt injection?
Injection, jailbreak and "note to the AI" patterns are detected by deterministic code before the model runs. For leads, the final decision is a rules gate, so even a model fooled into a perfect score can't push a lead through. The app's Security screen runs 13 such attacks live.
Where does my data live?
Either on your own server (we install the stack there) or on our hosted stack, where every row is tagged with your client id and isolated from other clients. Models are called through APIs; nothing is used to train them.
Which AI models do you use?
Through OpenRouter, the models that measure best on our own test sets, each with a strict deadline and automatic fallbacks: GPT-6 Luna for scoring leads, then DeepSeek V4.1 Flash, plus OpenAI embeddings for search and ElevenLabs for voice. Docket and Dossier read documents with Gemini 3.5 Flash-Lite first, then DeepSeek and Luna, and documents only go to providers that don't store or train on them. All three are provider-agnostic, so models can change without changing your configuration.
How long does a setup take?
Adding a business is configuration, not code: a profile, documents, optionally scoring rules and reply templates. The target for a standard setup is two weeks from the scoping call to live.
Why are the apps behind a password?
The apps run real models on every click, so they aren't open to the public. The numbers on this site are read live from the same systems, and each product has a public try-it page: score a lead, read your own invoice, or forge a synthetic applicant (erased after 30 minutes). For a guided walkthrough or a login, get in touch.
See them run on real data.
The apps run real models on every click, so they sit behind a password. Ask for a login or a guided walkthrough: talk to our AI receptionist in Front Desk, drop invoices into Docket, or load forged applicants into Dossier. Or try them in public: score a lead, read your invoice, forge an applicant, or connect Docket and Dossier to Claude. Or hire us to build yours.