Custom AI agents, shipped as real apps.
Your workflow, built into a custom AI agent your team actually uses.
Your choice of model and hosting — frontier or open-source, your cloud or ours in Switzerland.
Where AI gets useful
A chatbot in a browser tab can sound impressive — but it doesn't know your business.
AI gets useful when it works with your files, your data, and how your team already works — your tasks, on your documents.
And you stay in control: you see what goes in, you check what comes out, and nothing counts until you approve it.
The AI does the heavy lifting. Your team keeps the final say.
Your app sits in the middle — connecting the agent, your tools and data, and your team's judgment.
What you get — and what you don't
Fixed scope is the feature. You're buying a finished outcome, not a timesheet — so the boundary is written down before we start.
Included
- Your custom agent, built around your workflow — behind your login
- Screens to review and approve before anything counts
- Runs where you choose — your cloud or ours in Switzerland, frontier or open-source models
- Deployed, with a handoff and a short training
- You own the code — open-source base, no lock-in
Not included
- Data pipeline / warehouse work
- SSO / SAML / enterprise audit
- Billing or subscription systems
- Procurement / security-review cycles
- Open-ended R&D or "agent platform" builds
How it works
Intro call — 30 minutes, free
You show me the workflow. I map it — inputs, judgment points, outputs — and agree the scope. You keep the map whether or not we work together.
Set up & connect
Your app is running with your data and tools connected — the agent doing your first real task end to end.
Build & review
Your custom agent, with the screens to see inputs, review the output, and approve before it counts. You try it as it takes shape.
Deploy & hand off
Deployed where you choose — your cloud or ours in Switzerland. A short walkthrough, and it's yours.
Quotation
Is this you?
This is for you if
- You have a recurring workflow an agent should run — with or without a prototype today
- There's a demo, pilot, or internal rollout you want within 30–45 days
- Your data must stay private — your infra, your keys
- You need more than a chat box: files, tables, review, approvals
Not for you if
- There's no workflow yet — you want exploratory AI R&D
- You want a generic internal ChatGPT clone
Built in the open
Start here
Get in touch.
Tell me about one task you'd like an AI agent to handle — two sentences are enough. Within 48 hours I'll reply with what it could look like as a custom agent: the steps, the review, the result. Free, before any call.
We're already using Boring UI — can you customize it for us?
Yes — that's the core of what I do. I build your specific agent and workflow on the framework I maintain, then deploy it to your infrastructure. Customization by the person who knows it best, and you keep full ownership.
Why not build it ourselves with Cursor / v0 / Lovable?
You can — that's how most of the prototypes I see were built, and it's exactly where they stall. Those tools produce another demo. The sprint delivers the part they don't: auth, deployment on your infra, review surfaces, approval loops, telemetry, and a handoff your team can maintain. If your engineer has three free weeks, keep them; if not, that's the sprint.
Who owns the code?
You do. The foundation is MIT-licensed open source, the workflow plugin is written for you, and everything runs in your repos and your cloud. No license fees, no lock-in — hosting is optional and priced on demand, never a dependency.
How private is it — where does our data run?
Wherever you want it: your own cloud, or hosted by me in Switzerland. Use frontier models through your keys, or fully open-source models running locally for total sovereignty. Either way your data stays under your control — that privacy is most of why these agents exist, because serious work can't live in consumer chat tools.
Which models and stack?
Model-agnostic: Claude, OpenAI, Gemini, or local models — switchable per workflow. The stack is TypeScript, Postgres, and your cloud (or Fly.io/Vercel if you'd rather not think about it).
What if scope grows mid-sprint?
We agree the scope on the intro call — that's what keeps the fixed price honest. New ideas go on a list; we ship the agreed scope first, and anything extra becomes a second sprint.
About
Hey, it's Julien 👋
I'm a data engineer in Geneva. I've spent ten years building data platforms for companies — and lately I kept seeing the same thing: a clever AI demo that never turns into something a team can actually use.
So I built the open-source framework for exactly that, and now I use it to ship custom AI agents for teams. You work with me directly, start to finish — and you keep full control: your code, your data, your choice of model and hosting.
I write about building AI agents and data systems — roughly once a week, no fluff.
Read the newsletter on Substack ↗