There’s a question we get on almost every first call: which model do you use?
The honest answer is: whichever one you want. Today, a lot of our customers run on Claude, because it’s excellent at the analytical work finance teams need. Tomorrow they might run on Gemini, or GPT, or a model hosted entirely on their own servers. With Verain, that’s a setting, not a rebuild.
That’s a deliberate design choice, and it’s worth explaining why.
The single-model era is ending
For a couple of years, the default move was to pick one frontier provider and build everything on top of it. It was the fastest path, and at the time it was reasonable.
It’s also how you accumulate the most durable kind of lock-in. When your orchestration, your prompts, your tooling, and your team’s habits all assume one provider, “we could switch” quietly becomes “we never will.” And the provider knows it. They set the price. They set the terms. They decide what your data can and can’t be used for.
For mission-critical work (the numbers you take to your board, the contracts you sign), that’s a dependency, not a foundation.
Model-agnostic is the safer architecture
So we built Verain to treat the model as a component you can swap, not a wall you build against. Your Brain (the private intelligence we build from your data) sits behind an abstraction layer. Underneath it, you can run:
- a frontier model in EU-hosted cloud, when you want the best general reasoning available;
- a self-hosted open model on your own hardware, when control and cost matter more than the last few points of benchmark performance;
- different models for different jobs, because the best model for explaining a variance isn’t always the best model for drafting a memo.
Swapping is configuration. Your data, your context, and your workflows don’t move.
On-prem stopped being a compromise
The reason this matters more in 2026 than it did in 2024 is that the open models got good. On-premise open models now handle the large majority of enterprise workloads at a quality most users genuinely can’t tell apart from the cloud, and they do it at a fraction of the per-token cost. For a regulated business, “we run it inside our own walls” is no longer a quality sacrifice. It’s a defensible architecture you can put in front of an auditor.
What this means for you
You don’t have to bet your company on one vendor’s roadmap, pricing, or data policy. You bring the model you trust, run it where your rules require, and change your mind later without starting over.
That’s not a feature we tacked on. It’s the point.