How the pilot runs
The stages from discovery to the results session, what you need to provide, and how the work splits between what every estate inherits and what is yours.
How the 8 to 12 weeks run
- 00 Discovery Free. Define the workload, the data boundary and what success looks like.
- 01 Deploy and federate The stack runs in your environment and sign-in federates to your provider; federation to a commercial identity provider is in build.
- 02 Workload live Your people use it on real work, with policy and audit in force.
- 03 Admit and compare Each model in scope gets its provenance artefact, and candidates run against your own tasks, scored by an executable check.
- 04 Results Utilisation-aware TCO on your own tasks, measurements, documentation and a production recommendation.
What you need to provide
Accelerator capacity, a technical contact who can deploy in your environment, an identity administrator, a real workload with real data and the people who use it, and a decision maker for the results session.
What is pooled, and what is yours
-
Pooled work, inherited
Licence, provenance, checkpoint integrity and runtime compatibility, adjudicated once and inherited by every estate. Export screening is not yet assessed, and no export claim is made.
-
Marginal work, yours
Evaluation on your codebase and tool-call patterns, your cost per completed task, and the re-benchmark when the model or codebase moves. It is the per-model line on the price list.
Updates and security patching
Signed, content-addressed bundles you pull when it suits you. Nothing is pushed at you, and an air-gapped site receives the same artefacts by the route it chose.
Support
Business hours during the pilot, with a service commitment and out of hours cover at production tiers.
What the subscription carries after the pilot is on the pricing page.