Every AI model running inside Altara has a public model card.
What it is, what it's used for, what goes in, what comes out, where data flows, the human-in-the-loop checkpoints that wrap it, and how to report a concern. Updated when the underlying model, prompt scope or data-flow posture changes. No black boxes.
NAVI
The analytical layer behind every Altara module — drafts, scores narration, scam analysis, vision/OCR. Built on Anthropic Claude Sonnet 4.5.
Sentinel Transcribe
Converts uploaded voice notes to text so NAVI and the human analyst can review the content of suspected scam audio. Built on OpenAI Whisper-1.
A new card version is published when (a) the underlying foundation model is changed, (b) the system-prompt scope is materially changed, (c) new use cases are added inside Altara, or (d) the data-flow, residency or retention posture changes.
- General questions: hello@altaracore.ai
- Responsible-AI / incident: subject prefix
[ALTARA-RAI] - Right-to-deletion: subject
[ALTARA-DATA-RIGHTS]