1. Processing routes
DIRI AI uses model and voice providers only for a function selected in an authenticated training session. The session binds one training language and identifies the type of AI processing without exposing credentials or security-sensitive configuration.
| Provider | Function | Data needed |
|---|---|---|
| OpenAI | realtime or text simulation, transcription, analysis, scoring and coaching feedback | session instructions, selected scenario, necessary audio or text turns, transcript, scoring inputs and technical metadata |
| ElevenLabs | speech synthesis and, where expressly offered, voice processing for the selected language | generated reply text, audio where needed, language, voice/profile identifiers and technical metadata |
The exact model may change within the same disclosed purpose. DIRI AI does not route Customer Content to a deprecated, experimental or quality-assurance model as part of the paid service unless the customer-facing function and applicable safeguards expressly cover that route.
2. Provider roles and safeguards
DIRI AI uses a model provider as a processor for customer-requested functions where that role applies. Its Vendor and Transfer Register records:
- the contracting entity and applicable data-protection terms;
- whether content may be used to train provider models and the applicable account setting;
- provider abuse-monitoring and retention periods;
- processing locations and remote-support access relevant to the service;
- transfer mechanism and assessment;
- deletion and incident-notification controls; and
- voice licence/provenance and prohibited-use terms.
DIRI AI does not claim “zero retention”, “EU only” or “not used for training” unless account-level evidence and the applicable provider agreement support that claim for the exact route.
3. Languages and voice quality
Czech, English, Slovak, Polish and German may be offered for an interface or training path only where the selected product flow identifies that language. A provider may advertise more languages, but DIRI AI does not promise them unless the complete product flow, listening quality, accent, pronunciation, transcription, safety and support are available for that language.
Synthetic voices can vary. A voice trained primarily in another language may retain its native accent or become less stable. The selected training language is fixed for a session so routing and evaluation do not change silently mid-call.
4. Content use and retention
DIRI AI sends only data needed for the chosen function. Customer Content is not used to train a general-purpose provider model without a separate, voluntary and specific choice and lawful basis. Optional quality review is minimised, access-controlled and disclosed.
DIRI AI's own retention is in the Retention Policy. Provider copies may follow different short operational or abuse-monitoring periods identified in the applicable provider terms and Vendor and Transfer Register. Removing content from DIRI AI does not by itself prove deletion at a provider; the deletion workflow covers both where the relevant provider offers a deletion control.
5. Model changes
DIRI AI may replace a model for security, availability, quality, cost, deprecation or legal reasons. A replacement must satisfy the same disclosed purpose and applicable privacy, language, quality and security requirements before it receives Customer Content. A materially different provider, purpose, retention, transfer or decision capability requires updated notice and, where applicable, subprocessor notice or consent.
6. AI output and oversight
Model output can be wrong, inconsistent or biased. It is training feedback, not professional advice or an employment decision. The AI Transparency Notice describes review, challenge and prohibited uses.
Questions: support@diriai.com. Privacy questions: privacy@diriai.com. Version: 1.2.0. Date: 30 August 2026.