Fixed project•4–6 weeks
Hephaestus Sprint
Four to six weeks. One RAG system or one agent in production, with a golden eval set, monitoring hooks, a runbook and a go or no-go to scale.
See the package →
AI - AI that ships, in three packages
Three packages for AI that has to work after launch. A Hephaestus Sprint ships one system. Managed AI Ops keeps a live system healthy. A coding rollout sets up your engineering team. Private or VPC on AWS, Microsoft Azure, Google Cloud or GleSYS, or on-prem, by default.
One system
Tell us what should be in production
A workflow, a live AI system that nobody owns, or an engineering team without a shared way to use coding agents. We say which package fits.
Packages - Pick the shape of the work
Fixed scope before we start. You own the code, the prompts and the data.
How we deliver - What every AI engagement includes
The package changes. These do not.
- A golden eval set and a CI gate. Every AI engagement ships a set of real examples and a check that fails the build when a model or prompt change gets worse. We agree the set before we call it done.
- Your environment, by default. Private or VPC on AWS, Microsoft Azure, Google Cloud or GleSYS, or on-prem. Data stays in your environment, with audit logs. We do not train on your data unless you ask us to, in writing.
- A build, not a workshop. We write the code in your repo and leave the system running. We do not sell a strategy workshop.
How a package starts
We name the system, the boundary and the price before we build.
1. One system, named
Internal RAG, a customer or support agent, an ops agent, or the coding setup for one repo. If it is not one system, it is a different engagement.
2. Fixed scope, fixed price
You get the scope and the price before we start. Care after launch is a monthly agreement. New features are change orders, also at a fixed price.
3. In production, with a way back
Logging, permissions, the eval gate and a runbook ship with the system. You can turn it off.
How we work - We run on our own agents
The packages are the same operating discipline we use ourselves. Our portal, and the work in sales, recruiting and engineering.
- Our portal. The company runs on our own portal. Agents sit in the work, with permissions and a log of what they did.
- Sales and recruiting. Agents help the people who sell and the people who hire. The boundary is the one we put in a client system: least privilege, and a human when the step needs one.
- Engineering. Our engineers use repo context, skills and CI gates. The AI Coding Rollout is that setup, written so your team can run it.
Why teams ship AI with us
Senior engineers in Gothenburg. We build, and we stay for the care if you want that.
- Forward-deployed
- For automotive and connected products in Gothenburg, a senior sits with your team. The product engineering around the AI feature is part of the job.
- Builders
- You talk to the people who write the code. We work in your repo and follow your review rules.
- You know what it costs
- Fixed price for a sprint and for each change order. A monthly agreement for care. Nothing outside the agreement until you have approved it.
Also - A person in the team, or a platform
The packages are fixed scopes. Some teams want an engineer in the team. Some have a platform nobody owns.
Related insights - More on AI in production
Articles on taking AI from a pilot to something that runs.
FAQ - Common questions
Short answers on how the packages fit together.
Which package do we need?
One system to put into production: the Hephaestus Sprint. Something already live that needs an owner: Managed AI Ops. An engineering team that needs a shared way to use coding agents: the coding rollout. A person in your team for longer: consulting. A platform nobody owns: Platform Partner, then the AI upgrade path.
What does it cost?
You get a fixed price before we start. A sprint is one fixed price. Care is a monthly agreement. New features are change orders, priced one by one before we begin. We do not publish a rate card.
Where does our data go?
It stays in your environment unless we have agreed otherwise. Private or VPC on AWS, Microsoft Azure, Google Cloud or GleSYS, or on-prem, is the default, with audit logs. We do not train on your data unless you ask us to, in writing.
How do we know it works?
A golden eval set of real examples, agreed before we call it done, and a CI gate that fails the build when a change gets worse. After launch we watch quality, latency and cost.
What about the EU AI Act?
It depends on what the AI is used for. We help you see which category the system falls into, and we build in logging, traceability, documentation and human control. The legal opinion sits with your lawyer.
Who owns the solution?
You do. Code, prompts and data. We can keep it healthy on Managed AI Ops, or hand it to your team with the runbook.
Ready to talk?
Tell us where it is stuck
A demo that never shipped, a system nobody owns, or a team using coding agents without a standard. We say straight whether we are the right fit.
Let's talk about what you're building
