AI, language models and agent tooling
We build assistants, document pipelines and AI features with Claude, OpenAI, Gemini and Mistral AI, run open models privately with Ollama and Hugging Face, store embeddings in pgvector or Qdrant, orchestrate with the AI SDK, LangChain and LangGraph, connect tools through Model Context Protocol and automate with n8n and Zapier. For development teams we set up Claude Code, Cursor and GitHub Copilot.
- ClaudeOur default model for assistants and agents: strong reasoning, long context and careful with instructions.
- Claude CodeThe coding agent we set up for teams, working in the terminal and the editor on real repositories.
- OpenAIGPT models and embeddings through the API, for assistants, extraction and search.
- GeminiGoogle's models for long documents, images and video, used where multimodal input matters.
- Mistral AIEuropean models with strong multilingual quality, hosted or self-hosted.
- AI SDKVercel's toolkit for streaming AI in Next.js, with one interface across model providers.
- LangChainBuilding blocks for retrieval, tools and chains when an assistant needs more than one step.
- LangGraphStateful agents with branches, retries and human approval steps.
- Hugging FaceThe home of open models and datasets, for private deployments and fine-tuning.
- OllamaOpen models running locally or on your own servers, for data that cannot leave.
- QdrantA dedicated vector database for large document collections and fast semantic search.
- pgvectorVector search inside PostgreSQL, so retrieval lives next to your existing data.
- Model Context ProtocolThe open standard that connects agents to your tools, databases and documents, with permissions.
- CursorAn AI-first code editor we configure with project rules and shared conventions.
- GitHub CopilotCode completion and review agents inside GitHub, configured for your repositories.
- n8nVisual workflow automation your team can read and edit, self-hosted or in the cloud.
- ZapierQuick integrations between everyday business tools, without code.
- ElevenLabsNatural voices for assistants, read-aloud and voice interfaces in many languages.
Why it matters for your business
- Answers grounded in your businessRetrieval over your own files and database means the assistant knows your products, prices and policies, not just the internet.
- Your data stays yoursModels run under your accounts with training disabled, or privately on your infrastructure, with logs and embeddings in your systems.
- Costs you can predictModel routing, caching and budgets keep the bill proportional to use, and switching providers is a configuration change.
- Fits your existing productAI features are added to the app and workflows you already have, through the AI SDK and your own API, not as a separate tool.
Pick the model last.
The model is the most visible part of an AI project and the least important to choose first. What matters is the data it can see, how its answers are tested and what it is allowed to do. Get those right and most current models give good results; get them wrong and no model will.
So we start with your documents and real questions, build retrieval with pgvector or Qdrant, write a test set and measure. Then we choose between Claude, OpenAI, Gemini, Mistral AI or an open model through Ollama based on quality, cost and where the data may go.
The tooling around the model is chosen for longevity. The AI SDK gives us streaming and provider switching inside Next.js, LangChain and LangGraph structure multi-step agents when a task needs them, Model Context Protocol connects agents to your tools with permissions you control, and n8n or Zapier handle automations your team can see and edit. For development teams we configure Claude Code, Cursor and GitHub Copilot with written guidelines, so the same discipline applies to the code that builds all of this.
How we work with it
- 01
Audit and plan
We review what you have, define the goal and write a proposal with scope, timeline and cost.
- 02
Design and prototype
Structure and screens in Figma, approved by you before any code is written.
- 03
Build in sprints
Short cycles with a demo at the end of each, on a staging environment you can test.
- 04
Launch and support
Deployment, monitoring, training and a backlog for the next phase.
Related services
SaaS developmentSubscription products built to grow: accounts, billing, roles and an admin panel, on a stack your next developer will know.
AI developmentAssistants, document processing and AI features inside your product, built on Claude, OpenAI and open models, with your data kept yours.
AI for development teamsAI-assisted development set up for your team: Claude Code, Cursor and Copilot configured, guidelines written and your codebase made agent-ready.
AI automationBack-office work done by software: emails sorted, documents processed and data moved between your tools, with a person in the loop where it matters.
Questions and answers
All technologies
FrontendInterfaces in React, Next.js, Vue and Angular.
BackendAPIs, authentication and the logic behind the screen.
DatabasesThe right store for every kind of data.
Cloud and DevOpsHosting, deployment, containers and CI/CD.
Tooling and workflowBuild tools, linting and monorepos that keep code healthy.
Testing and QAAutomated tests that catch problems before customers do.
Mobile appsiOS and Android apps, native or cross-platform.
Design and collaborationDesigned in Figma, planned in the open.
Monitoring and dataSee errors, performance and events as they happen.
Analytics and growthMeasure what matters and keep in touch.
Payments and e-commerceOnline shops and payments that just work.
Content managementEdit your own site without a developer.
Get a quote for your AI project
Every project is quoted individually after a short call. Tell us about your business and you will get a written proposal with scope, timeline and cost within one working day.
- 1We reply within one working day
- 2A short call to understand your goals and what you have today
- 3A written proposal with scope, timeline and cost, no obligation
