Zeus AI Quality Assurance Overview
Last updated September 17, 2026
The QA feature is where you can test how Zeus AI behaves before you take it live. Use this section to try real customer scenarios, see how Zeus AI responds, and understand how it uses your Knowledge, Tools, Protocols, Guardrails, and Tone — all without affecting what customers see.
Key features
- Test mode: A real-time sandbox to simulate customer prompts against your live or overridden Knowledge, Tools, Protocols, Guardrails, and Tone — test with specific customers, file attachments, and even different models before anything goes live.
- Evals (Internal info): Build a dataset of customer messages, ideal responses, and expected tool calls, then run individual or bulk tests to catch regressions before a protocol or knowledge change reaches customers.
- Run history: A searchable log of past conversations with review states (Pending, Reviewed, Flagged), plus tabs for the exact knowledge retrieved, protocols applied, tools triggered, and the AI's reasoning behind each response.
Best practice for you
Treat QA as a loop, not a one-time check: test a change in Test mode, and review Run history on a regular cadence (weekly works well) to flag anything that slipped through. That habit is what keeps response quality steady as your Knowledge, Protocols, and Tools keep growing.
Important: For more information, please refer to the following resources: How to manage Zeus AI: QA - Test mode How to manage Zeus AI: QA - Evals (Internal) How to manage Zeus AI: QA - Run history