AI inside your web product
AI that does its work inside your web app.
We add AI features to the web product you already run — search that answers, forms that fill themselves, assistants that know your data — or build a new AI-first product end to end, from interface to infrastructure.
An example help page inside a web app: a visitor asks how to move their billing date, and the answer streams in with numbered citations that point to two help articles.
What it is
Most AI features fail in the interface, not the model: an answer that arrives too slowly, a citation nobody can check, no plan for when the model is unsure. We design the feature and the plumbing together — streaming responses, sources people can open, sensible limits on cost — and build it into the stack you already have.
This page is one part of AI inside your product. The same service covers Android and iOS apps.
The spec
- What you get
- An AI feature built into your existing web app, or a new AI-first product
- Streaming interfaces, citations, and graceful fallbacks when the model is slow or unsure
- A server layer that keeps keys, prompts and spending limits off the browser
- Monitoring for quality, response time and cost once it is live
- How we measure it
- Task success for real users, time to the first words and to a finished answer, and cost per active user — tracked in production against the bar we agree before we build.
- A good first step
- Pick the feature your users ask for most. We prototype it inside your app, on your data, and test it against real requests.
What we build with
Our usual stack for web work. If yours is different, we work in yours.
- React
- Next.js
- TypeScript
- Node.js
- Python
- PostgreSQL
- pgvector
- Streaming responses
- REST and GraphQL APIs
- Vercel
- AWS
- Docker
Where it fits
Examples of what this work looks like — not client stories.
- Example
Search that answers
A help centre or docs site where people ask a question and get a cited answer instead of ten links.
- Example
Forms that fill themselves
Onboarding, quotes or long forms that fill in from a document or a short description, with every field still editable.
- Example
An assistant in your dashboard
People ask their own data questions in plain language and get a table or chart back, with the query shown.
- Example
A new AI-first product
From an idea and a dataset to a working product: interface, model choice, evaluations and hosting.
Common questions
Yes — that is most of this work. We start by reading your codebase and how the feature would fit, then build inside your conventions instead of bolting on a separate app. Your team reviews and owns the code.
Whichever fits the job on quality, speed and cost, measured on your own requests. We keep the model behind a small interface, so it can be switched when a better or cheaper one arrives.
Requests go through your servers, not straight from the browser to a model provider. We use providers that do not train on your data, or run open models in your own cloud when that matters.
It depends on the feature and on your codebase, so we do not quote a number before we have seen either. We agree the scope of a first prototype on the discovery call and estimate once we have read the code.
We can keep watching quality and cost in production and keep improving the feature against the same bar. That is our Run & improve service.
What should AI do in your web app?
Tell us the feature and who uses it. We will come back with how we would build it, what it would cost to run, and how we would measure it.