For software vendors
AI features in your product, on models you control.
We choose and test the right open-weight models for your product, run them in your own AWS cloud and operate them for you. Your team stays on the product.
- Cloud
- Yours, in the EU region you choose, for example Frankfurt.
- Models
- Open-weight models, chosen and tested on your product’s tasks, for example from the Llama, Mistral, Qwen or Gemma families.
- Runs on
- Amazon Bedrock, the managed AWS model service, or dedicated GPU instances, whichever fits each model and your workload.
- Languages
- German and English, tested on your own examples.
- Fees
- A fixed setup fee, then a monthly operations fee, agreed in writing. AWS bills your usage directly to your own account.
01 / What we do
Everything between your product and the model
You decide which AI features your product should have. We find the models that do them well, run them in your cloud and keep them running, so your engineers can stay on the product.
Model selection and evaluation
There are hundreds of open-weight models, and they differ in language skills, reasoning, speed and running cost. For each feature we shortlist candidates, build a test set from your product’s real tasks and compare them on it. You decide with the results in front of you.
Pinned versions
Every model runs at a version you know. On the dedicated-GPU setup it stays fixed until you approve a change, so your prompts, your tests and your product’s behaviour do not shift under you.
Inference in your cloud
Your product calls a private endpoint inside your own cloud. Behind it runs Amazon Bedrock or a dedicated GPU instance, whichever fits the model and the workload.
Scaling with your product
We plan capacity for the traffic you expect and adjust it as your customer base grows, so the AI features keep pace with the rest of your product.
Monitoring
We watch health, errors, capacity and cost, and step in when something drifts. Your team does not have to keep an eye on an inference stack.
Upgrades
Better open-weight models keep arriving. We test the promising ones on your test set, show you what would change, and switch only after you approve.
Model catalogue
The open models that matter, running in your cloud.
Open weights mean the whole model runs inside your cloud. Your prompts never reach the company that trained it, wherever it comes from. We test the candidates on your own examples, check each licence for your use, and keep the catalogue current as new releases appear.
- DeepSeekDeepSeek
- QwenAlibaba
- GLMZhipu AI
- KimiMoonshot AI
- LlamaMeta
- MistralMistral AI
- GemmaGoogle
- gpt-ossOpenAI
- and many more
Model names and logos are trademarks of their respective owners. No affiliation or endorsement is implied.
02 / Your own cloud
Why your own cloud matters, for you and your customers
If you sell to law firms, tax advisers, banks, insurers or the public sector, their questions about AI become your questions. With the models in your own cloud, you have clear answers.
Their data never goes to an AI vendor
Your customers’ prompts and documents are processed inside your cloud and are not sent to any AI model vendor. There is no AI vendor in the data path for you to explain, assess or answer for.
You can show where it runs
Region, services, access: all of it sits in your cloud, so you can answer security questionnaires and customer audits with specifics. Our access is time-limited, logged in your own CloudTrail, and yours to revoke at any time.
Versions do not change under you
On the dedicated-GPU setup, the model version stays fixed until you approve a change. What you tested, documented and promised your customers is what keeps running.
The paperwork is part of it
We sign a GDPR processing agreement and a separate secrecy agreement with you, and help with what your customers will ask for: an AI notice for users, a licence register for the models and input for a data protection impact assessment.
03 / How we work
From first call to production
Assessment
Thirty minutes on your product, the AI features you have in mind, your customers’ requirements and your AWS setup. You leave with a concrete recommendation for the next step.
Proof of concept on your data
We shortlist models, build a test set from your real examples and run the candidates in your cloud. You see how each one handles your tasks, in German and English, before anything goes near production.
Production
The chosen setup goes live in your cloud: a private endpoint for your product, capacity planned for your traffic, access you control, and the agreements signed.
Operations
We keep it running: monitoring, updates, capacity, and model upgrades you approve. You have a named person who knows your setup, with support during business hours (CET).
04 / In writing
What we promise
Specific statements, not slogans. Two of them depend on the setup, and we say which.
Who can see what, in detail- Runs in your own cloud. You own the keys and can revoke our access at any time.
- Your prompts and documents, and your customers’, are not sent to any AI model vendor.
- Hosted in the AWS EU region you choose, for example Frankfurt.
- Our access is time-limited and logged in your own CloudTrail.
- On the dedicated-GPU setup, the model version stays fixed until you approve a change.
- On the dedicated-GPU setup, AWS commits in its Service Terms that AWS staff have no technical means to access your content on Nitro EC2 instances.AWS Service Terms
- A GDPR processing agreement and a separate secrecy agreement, signed with you.
- Help with the paperwork: an AI notice for users, a licence register and input for a data protection impact assessment.
Tell us what your product should do with AI
Thirty minutes on your product, your customers and your AWS setup. You leave with a concrete recommendation: which models to test, how to run them, and what operating them takes.
Book an assessment