We design, build and run private AI systems for business — chat, voice and agents that work like your team, on hardware you control, with your data never leaving the building.
Small, purpose-built models that score and decide rather than chat. Given a state and a fixed set of options they return a choice with calibrated probabilities in one forward pass — no prose, no sampling, and a confidence you can threshold on.
| model | what it decides | formats |
|---|---|---|
| amaara-decide-vision | screenshots, charts, diagrams | bf16 · FP8 · GGUF |
| amaara-decide-business | triage across four business workflows | bf16 · FP8 · GGUF |
Both 0.8B, run on CPU, with per-bucket calibration baked into the artefacts.
A model that scores well can still be useless if it is not reading its input. Every card reports the check that would catch that:
Where a model falls short, the card says so.
An agent for every surface — IDE, CLI, Desktop, and Mobile (coming soon). Point it at any model you host: cloud, local or air-gapped. Made and maintained in Australia. Also in the ecosystem: Open CRM · Open Jobs · AI Commentator.