KYRA · Trust Center
AI Transparency & Acceptable Automation
KYRA is an AI product, so you deserve a plain account of how it uses AI, what happens to your data, and the controls you hold.
Version: 1.0 · Effective: 23 August 2026 · Owner: Kyra.bot Ltd
1. You are interacting with AI
KYRA is an AI-assisted IT-support service. In Teams, Slack, SMS, phone and web chat, replies may be generated or materially assisted by an AI system. The phone greeting expressly identifies KYRA as an AI assistant.
This disclosure is provided for transparency, including where Article 50 of the EU AI Act applies.
2. What AI does
Depending on Customer configuration, KYRA may:
- retrieve tenant-scoped knowledge;
- answer and summarise support requests;
- classify urgency, impact or ticket fields;
- create or enrich PSA tickets;
- identify knowledge gaps;
- review historical ticket quality and suggest coaching;
- retrieve selected Microsoft 365 information;
- propose Microsoft 365 actions;
- execute a narrowly configured action only after trusted server-side authorisation and any required approval.
3. What AI does not guarantee
- AI can misunderstand a request, use outdated or incomplete knowledge, or produce an incorrect answer. An Output is not guaranteed to be accurate, complete, unique or suitable for a particular purpose.
- Do not rely on KYRA for emergencies, physical safety, or medical, legal or financial advice, or for irreversible or high-impact action without competent human verification.
4. Knowledge and providers no training
- By default, AI inference and embeddings run on Microsoft Azure OpenAI in the UK (UK South). Answers may use Customer-supplied knowledge, conversation context and connected-system data. Enterprise customers may pin a different provider or region, or bring their own (OpenAI, Anthropic, or a self-hosted model). Current recipients and regions are listed in the Privacy Notice and Subprocessor Notice.
- Customer Data is not used by KYRA to train a general-purpose model for other customers. Provider handling is governed by the applicable business or API terms and configured data controls.
5. Human assistance
- End Users can ask for a human or request escalation. Availability and response time depend on the MSP or End Customer support arrangements.
- A failed or refused automated action should be escalated through the configured support path.
6. Microsoft 365 actions
Action capability is configured per customer and action type. A proposed action is not authorised merely because an AI model suggested it. KYRA re-checks identity, scope, allowed targets, capability mode and approval state in trusted code. Modes may include:
- Disabled.
- Information or proposal only.
- Human approval required.
- Automatic only for explicitly approved, low-risk, scoped actions.
Customers must choose modes appropriate to risk and law. KYRA is not intended to make solely automated decisions producing legal or similarly significant effects on an individual.
7. Ticket QA and people data
Ticket-QA features may evaluate ticket records and generate quality or coaching suggestions about technicians. These are decision-support outputs, not definitive employee-performance determinations. Customers must provide workforce notices, assess fairness and permit human challenge or review before using them for employment decisions.
8. Feedback and correction
Users should report inaccurate or unsafe replies through the available feedback or escalation mechanism. KYRA or the Customer may use that feedback to correct knowledge and improve the configured service, subject to the Privacy Notice and DPA.
← Back to portal