LLM integrations
Language models wired into your existing product or stack, with cost and quality controls.
- Engagement
- Fixed-scope engineering, usually inside an existing product
- Typical timeline
- 4 weeks
- Starts with
- A read of your current model calls and last month’s provider invoice.
The feature works. It is wired straight into one provider, last month’s bill tripled, and nobody can say which customer caused it. When that provider has an incident, the feature has one too, because there is nothing behind it.
AI features shipped without rebuilding what you already have.
What we build
Concrete artefacts, handed over and documented.
One internal interface for language model calls, so changing provider is a configuration decision rather than a rewrite
Routing by task: the cheap fast model for sorting and tagging, the careful expensive one for the paragraph a customer will read
A fallback chain that survives a provider incident, with the quality trade-off of each fallback written down beforehand
Spend caps per customer, per feature and per day, enforced before the call rather than discovered on the invoice
Prompt versioning, so a wording change ships like code and can be rolled back like code
What changes
Model spend becomes a line you can attribute to a feature and an account
A provider outage degrades the feature instead of removing it
Trying a newer model becomes an afternoon rather than a project
How it runs
- Week 1
01Audit calls and cost
We find every place your code talks to a model and reconcile it against last month’s invoice.
- Weeks 2–3
02Abstraction and routing
One interface, task-based routing, and a quality bar each route has to clear before it is allowed to serve traffic.
- Week 4
03Caps and fallbacks
Budget limits, the failover chain, and a dashboard showing spend by feature.
Chosen per project. Named here so you can see the shape of it.
- Vercel AI SDK
- OpenAI
- Anthropic Claude
- Google Gemini
- Langfuse
- Postgres
Questions we get asked
Tell us the outcome, not the tooling.
Send us the situation you are in. We will tell you which discipline it belongs to, what we would do first and what it costs, including when the answer is to wait.