The platform
Your Brain. Your model. Your data.
Verain builds a private Brain from your organisation’s own data, then runs specialised Minds on top of it. You decide which model powers it, and Verain trains no model on your data.
The Brain
The Brain knows your business.
The Brain is the layer that makes everything else useful. It connects to your systems and grounds every answer in your own data. For Finance today that means your ledger, your entities and your history, read from Exact Online, Twinfield, Odoo, Tripletex and Visma.net; Verain Brand, in development, adds your documents and your voice. It learns from your corrections, so accuracy compounds with use rather than resetting with every prompt.
Two ways to build a Brain, depending on the work:
Grounded.
Your data plus the model of your choice. Fast to stand up, model-agnostic, and the right fit for analytical work like Finance, where the value is in connecting to your systems and reasoning over live numbers.
Fine-tuned and owned.
For knowledge-heavy work like Brand, a private AI grounded in your own material and owned outright, deployed in your infrastructure. Optionally fine-tuned on your work so it writes and thinks the way you do.
Any model
Bring any model. Change your mind later.
The single-model era is over. Locking your most important work to one provider is a risk you don’t need to take. Verain runs on frontier models (Claude, GPT, Gemini), European models like Mistral, and self-hosted open models alike, and inference can stay in the EU through your own cloud account. Switching is a configuration change, not a re-platforming project.
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, at a fraction of the per-token cost.
Where it runs
Hosted in Germany. Your model, your route.
Verain runs hosted in Germany, with each tenant’s ledger data in its own database files. Model inference follows your choice: the default hosted provider, a route that keeps it in the EU under your own cloud account (AWS Bedrock in Frankfurt, Google Vertex in the EU, Mistral), or a self-hosted open model. The EU AI Act now arrives in two speeds: the Article 50 transparency obligations have applied since 2 August 2026, while the heavier high-risk regime was deferred to 2 December 2027 by the Digital Omnibus. Either way, “where does our data go” stops being a question you have to hope about and becomes one you can answer precisely.
Model
Frontier, European like Mistral, or open and self-hosted. Switching is a setting.
Runs in
Your data
- Isolated to your tenant, always
- Stays yours: Verain trains no model on it
- Stored in Germany
No black box
Transparent by design.
Verain is model-agnostic. The hosted default is Anthropic’s Claude, and you can choose another frontier model, a European model such as Mistral, or a self-hosted open-weight model, so there is no foundation-model lock-in. Access is governed by granular, role-based permissions. The architecture is auditable, so a regulated business can verify it rather than take it on trust.
The Minds
One Brain. Many Minds.
Every Mind inherits the context the Brain already holds, so each new vertical starts ahead instead of from zero.
See it on your own data.
The fastest way to understand the platform is to watch it work on real numbers. Start with Finance.