Placeholder AIWhere tbd-0 fits.
Typical situations where a smaller, sovereign model is the better choice.
A customer-facing agent with predictable cost
Situation. Your support or sales agent handles thousands of conversations a day. With token-based pricing, every turn of a long conversation costs more than the last, and the monthly bill is hard to forecast.
With tbd-0. Constant context and per-turn billing keep the cost of each turn flat. You know what a conversation costs before it starts.
On-premise AI for a security-critical application
Situation. Defense, critical infrastructure or internal security tooling. The data can’t leave your network, so a cloud API is not an option.
With tbd-0. We deploy the model on your own hardware. Our models are small and run on AMD and Intel hardware, so you don’t depend on scarce GPUs.
Private data that must not go to the US
Situation. Personal, health or confidential business data that you can’t hand to a provider under US jurisdiction.
With tbd-0. Our models run on European infrastructure with no US subprocessor, so the US Cloud Act doesn’t apply. We don’t train on your data.
An expert model for machine manufacturing
Situation. Technical documentation, service reports, specifications. General-purpose models don’t know the terms of your field and get details wrong.
With tbd-0. tbd-0-mm is trained for the language of engineering and production. It knows the field, so it can be smaller and still more precise.
A coding assistant that keeps your code in Europe
Situation. Your developers want AI help, but your source code can’t go to a US provider.
With tbd-0. tbd-0-cs is built for coding and software development and runs in our European cloud or on your own servers.
Long conversations and long documents
Situation. Contract reviews, research or support threads that run for hours. With a growing context, models get slow and lose details in the middle.
With tbd-0. The context stays the same size, so there is no length limit and nothing gets lost.
A different use case?
Tell us about it. If a general-purpose model isn’t enough, it might be our next expert model.