Services · 04
AI
AI support agents, chat over your documents and data, and automations that take routine work off your team. Built into your product, connected to your systems, and added only where it brings real value.
- What we build
- AI agents, assistants, AI features and automations
- Works with
- Your documents, databases and tools
- Models
- Anthropic, OpenAI and open-source models
- Team
- Senior-led, in direct contact with you
What we build
AI that does
Not a chatbot bolted on the side: AI inside your product and the workflows your team already uses, judged by the work it takes off people.
Support agents
Answer customers from your orders, docs and policies, and hand over to a person with the full context.
Chat over your documents
Ask questions across contracts, manuals and knowledge bases, and get answers with links to the source.
Workflow automation
Sort tickets, draft replies, fill in the CRM and route requests, with a person approving where it matters.
Document & data extraction
Turn invoices, forms, emails and photos into structured data your systems can use.
AI features in your product
Smart search, summaries, recommendations and writing help, in the screens your users already know.
Vision
Recognise what is in a photo or a scan: products, documents, damage, a plate of food.
Insights from your data
Ask your data questions in plain language and get reports, charts and alerts instead of SQL.
Agents with tools
Agents that look things up and take actions in your systems, within limits you set and with every step logged.
Built in
What makes AI
A demo takes a day. A feature people can rely on needs the parts around the model.
Your AI feature
A model behind your API, grounded in your data.
- Your data, connectedDocs, databases, tools
- RetrievalAnswers with sources
- Structured outputsData your code can use
- Tool useActions in your systems
- EvalsTested before every change
- GuardrailsLimits, filters, approvals
- Human in the loopReview where it matters
- PrivacyData controls, EU hosting
- Cost controlCaching, the right model per task
- SpeedStreaming, a fast first answer
- MonitoringQuality, cost, usage
- Any languageMultilingual by default
Stack
The right model
We choose models by the job: the strongest where quality decides, smaller and cheaper ones where they are enough, and your own hosting when data must stay in-house.
- 01
Models
Who does the thinking
- Anthropic Claude
- OpenAI
- Open-source models
- Vision models
- 02
Retrieval
How it finds your facts
- PostgreSQL · pgvector
- Embeddings
- Hybrid search
- Document parsing
- 03
Orchestration
How it works with your code
- TypeScript
- Python
- Model Context Protocol
- Structured outputs
- 04
Quality
How we know it works
- Evals
- Tracing & logs
- Human review
- 05
Delivery
Where it runs
- Node.js
- FastAPI
- Docker
- AWS

From our work
AI in a
One of the projects in our cases is Nimbus, an app for a private fitness club. Members take a photo of their plate or type a sentence, and AI counts the calories: a vision model names each dish and its portion, and the numbers come from public nutrition databases.
The model sits behind an API and returns structured results with a confidence for each dish, and its answers are tested with evals. Photos are used for the analysis and not kept, and every number is labelled as an estimate.
- Vision AI
- Structured outputs
- Evals
- iOS & Android
- EU hosting
Where could AI
Tell us which work takes your team the most time. On a free intro call we'll tell you honestly where AI can take it on, and where it can't.