← All solutionsSolution · AI assistants and copilots

An assistant that is right, safe and affordable

A chat, copilot or support agent inside your product, on your site or on WhatsApp. It answers from your own data, takes real actions like order lookups and bookings, hands over to a person when it should, and costs a known amount per conversation.

Who it is for

Consumer products adding a chat, an in-app copilot or a support assistant, and teams whose first one is confidently wrong.

What we can build for you

Six things we can put in your product, each backed by work we have already shipped.

01
In-app copilot
Answers questions about the user's own data from tools that read it, never from memory.
02
Sales and support assistant
On your site and on WhatsApp: product questions, order status, returns and recommendations.
03
Actions with guard rails
Order lookups behind a one-time code, cart changes, bookings and lead capture. A side effect never runs twice.
04
Human handover
A person can take over live, and the AI steps back while they do.
05
Analytics that talk back
Ask what changed this week and get an answer built from your dashboards' own numbers.
06
Conversation insight
Every closed chat classified, summarised and mined for leads and for demand you could not meet.

The four tests, for this

Where this kind of feature usually breaks, and what passing each test looks like.

01 · Design
Does it feel right?

Replies stream in, look like your product, and a person can step in at any moment.

02 · Mobile and backend
Does it hold up?

Holds a traffic spike, and never repeats a side effect when a call is retried.

03 · AI, retrieval and evals
Most at risk
Is it right?

Searches before it answers, scopes data per customer, and checks its numbers against what it read.

04 · Metering and pricing
Does it pay?

Every call metered, the prompt laid out for caching, small models wherever they are good enough.

Closest track: Build it from zero

How the two weeks go

€6,000, fixed. Written findings on day five; stop there and pay half. You keep everything either way.

  1. Days 1 to 4Read your data and your failure cases. Map which questions need retrieval, which need tools, and what each turn costs.
  2. Day 5Written findings: the architecture, the cost per conversation at your volume, and which of the four tests are at risk.
  3. Days 6 to 9A working assistant in your repo on your real data, with retrieval, the first tools and metering in place.
  4. Day 10Handover: the branch, an eval set built from your questions, the cost model and a two-page architecture note.

Where we have done it

Previous work · 2025 to now
Multi-agent AI platform
Twelve tools, hybrid retrieval, metering and 157 eval cases, in production with paying customers.
Built by our team · 2026, early access
AI product analytics platform
An assistant with seven read-only tools that answers only from what they return.

Tell us what you are building.

Thirty minutes, no deck. You get a straight answer about whether we are the right people, and we say no when we are not.