An AI that talks, and deterministic code that decides — the model can never bend a rule.
💬 The AI talks
Groq GPT-OSS 120B (or Claude / Gemini / keyless fallback) understands you and chooses what to do.
🔧 Tools act
Nine tools are the only way anything happens — verify identity, rebook, refund, compensate, escalate.
§ Policy engine decides
Every tool call is checked against the rules in code. Not allowed → it simply doesn’t execute.
🧑⚖️ Humans approve the rest
Beyond-policy asks become tickets a supervisor approves or denies — the answer lands back in your chat.
1 · You see it instantly
A ticket card appears in the chat and on your dashboard with its ID.
2 · A human decides
The supervisor reviews it in the Resolution Console (/admin).
3 · Aria tells you the outcome
The decision flows back into your chat within seconds — approved waivers even execute the rebooking automatically. Every message and action is kept in a permanent case record you can download.