Data used by the AI Agent
The AI Agent works with two categories of data: the Customer Content you provide and the Usage Data that OneSignal’s systems generate. For how these categories fit into OneSignal’s full data model, see Data taxonomy.What you provide (Customer Content)
The AI Agent processes data you provide, which may include:- Prompts and messages you type in the AI Agent interface
- Your campaign content, audience segments, and notification templates accessed through your OneSignal account
- Context from external systems you connect through MCP (Model Context Protocol) connections
- Skills Files and Brand Center assets you upload to guide AI responses
What OneSignal’s systems generate (Usage Data)
OneSignal’s systems also generate operational telemetry that does not contain your content, including:- Infrastructure and API performance metrics
- AI feature usage statistics (non-identifying)
- Engagement predictions derived from aggregated, de-identified platform data
Data retention
| Data type | Retention period | Notes |
|---|---|---|
| Agent Conversations | 30 days | Auto-deleted after 30 days. Deleted immediately on account closure or on customer request to privacy@onesignal.com. |
| AI evaluation and annotation logs | Up to 90 days | Retained in OneSignal’s Google BigQuery environment (europe-west4). May be reviewed by authorized OneSignal personnel for service quality purposes. Auto-deleted at or before 90 days. Not subject to individual deletion requests; see your DPA for details. |
| Skills Files | Until deleted | Deleted immediately when you remove them. Not retained by the AI model. |
| Brand Center assets | Until deleted | Deleted immediately when you remove them. Not retained by the AI model. |
| GEAP request/response logs | Not retained | Google’s GEAP request-response logging is disabled for OneSignal. Prompts and outputs are not stored at the Google infrastructure layer. |
Your controls
Disable AI for your organization
Organization Owners can disable all AI Agent features for an entire organization from your Organization Settings. This setting:- Applies to all users in your organization immediately
- Persists across OneSignal product updates. OneSignal will not re-enable AI features without your action
- Can be managed in your Organization Settings
Delete conversation history
You can request deletion of your Agent Conversation history at any time by contactingprivacy@onesignal.com. Conversations are also automatically deleted after 30 days.
AI evaluation and annotation logs are subject to a separate 90-day retention period and are not individually deletable. See the Data retention table above.
MCP connections
If you use MCP connections, you control which external systems the AI Agent can access. You can revoke access to any connected system at any time from your Organization Settings.What OneSignal does and does not do
OneSignal will not
- Use your Customer Content (prompts, outputs, campaign data, subscriber data) to train or fine-tune any AI or machine learning model
- Share your prompts or AI outputs with Google or any third party for training purposes. Prompts stay within Google’s GCP infrastructure
- Disclose your data externally in identifiable form
- Re-enable AI features you have disabled
OneSignal may
- Use aggregated, de-identified platform operational data (“OneSignal Analytics Data”), derived from platform operational telemetry only and never from your Customer Content or Subscriber Data, to improve AI features and platform performance
- Retain Agent Conversations in OneSignal’s own systems for up to 30 days to support conversation history features
- Retain AI Agent evaluation and annotation logs for up to 90 days for service quality purposes. These logs may be reviewed by authorized OneSignal personnel to evaluate and improve the AI Agent feature. These logs are not used to train any AI or machine learning model
How the AI Agent works
The AI Agent uses a retrieval-augmented generation (RAG) architecture:- Skills Files and Brand Center assets are retrieved per session to provide context. They are not fed into the model permanently or used to fine-tune it
- Conversation history is maintained in OneSignal’s own systems, not by the AI model provider
- AI inference (the actual language model processing) is performed by Anthropic’s Claude, accessed through Google’s Gemini Enterprise Agent Platform
Subprocessors
The subprocessors used in connection with the AI Agent feature are listed at OneSignal’s list of subprocessors.Data Processing Agreement
Use of the AI Agent feature is governed by the OneSignal Data Processing Agreement (DPA). Agent Conversation data is Customer Personal Data under the DPA. Google’s GEAP and BigQuery are listed as subprocessors. For questions about data processing, contactprivacy@onesignal.com.
Related pages
Data taxonomy
How OneSignal classifies data, who controls it, and what governs each category.
Handling personal data
How OneSignal handles personal data and how you can meet GDPR and CCPA requirements.