The small model for sovereign AI.
It understands the language in your data to find, compare and decide better. Designed for offline environments with sensitive information.
- CPU
- offline
- on-device
- 100% private
- Spanish and English
Sandbox
Try it live.
Ask what to do or say what changed, in your own words: tardi does the calculation with the scene’s figures and yours, decides and backs it with its document. If it is not sure, it does not decide.
Decision
In this demo, Model Foundry generated the mining 0.1 model from a model with more than 100M parameters and concentrated the capacity a specific domain needs into just 9.15M.
The model was trained for several common mining scenarios. It understands the question and tags each figure with what it measures; deterministic rules do the calculation, never the model. In a pilot, the situations, rules and documents come from the operation itself.
- model size
- 6.9 MB
- cost per query
- $0
- bytes sent out
- 0
- GPU required
- none
Runs on
- Intel/AMD PCs and laptops
- Apple Silicon
- ARM servers
- Raspberry Pi
- iOS and Android
First deployments
Bring tardi into a real operation.
We are looking for organizations with critical assets, disconnected operations or information that cannot leave their network.