Is this for you?
Straight answers to the questions most teams ask before a first call.
- What is a discovery session?
- A short working session on your highest-value AI workload, the data that cannot leave your environment, and what a successful pilot would need to prove. It is free.
- Is the pilot paid?
- Always. Discovery is free, the Estate Pilot is a fixed scope over 8 to 12 weeks priced to cover its own cost, and we do not run a free production pilot.
- What if the workload does not fit?
- We say so in discovery. If a cloud provider's private deployment already satisfies the requirement, or your volume is small enough that a cloud subscription costs less, we will tell you that too.
- Privacy is not our problem. Our AI bill is. Is this for us?
- Often, yes. Private AI swaps a per-token bill that grows with use for a cost you can predict. Compare measures, on your own tasks, whether it is cheaper at your volume.
- Can I keep my coding tool?
- Yes. Point it at a private OpenAI-compatible endpoint instead of a cloud provider: a new base URL and key, not a new tool. We do not build a coding agent.
- Does this work with a hosted provider or a frontier API, both in build?
- Both routes are in build, and in pilot scope when available. You enable a route in your own configuration; until then, nothing leaves. A hosted route, in build, is labelled provenance by host attestation, not verified. See the capability status table.
- What hardware do I need?
- We specify what the workload needs. You buy it, rent it, use a private data centre or use capacity you already have.
- How long does the pilot take?
- 8 to 12 weeks, from deployment to the results session.
- What do I get at the end?
- A working private workload, the deployment as code, security and data-flow documentation, model evaluation on your own tasks, and a production recommendation.
Who this is for
Organisations with a concrete reason a workload should not go to an external AI provider. The reason is usually one of two.
-
The data cannot leave
Source code, customer records or client documents that a contract, a customer or your own security team will not allow into someone else's environment.
-
The bill cannot keep growing
Per-token pricing has no ceiling, and agentic work makes it worse: the more useful a workload becomes, the more it costs.
Typically a handful of models across one or two environments, with a platform team able to run what we deploy. The first workload is usually private coding agents.
This is probably not for you if
You have one model in one environment, and nothing queued to change it. The admission question only arises once the estate starts to fragment.
Your workload is small and bursty, so a cloud subscription will likely cost less, and cost is not the reason you are looking.
You have no internal platform or infrastructure capability.
Your requirement is already satisfied by a cloud provider's private deployment. Most are.
You need contractual indemnity for model outputs.
We would rather say so in the first conversation than in week five.
Still have questions after the answers above? Heliast runs this inside your environment, on your policy.
Book a discovery session