Data Privacy and Retention
Alteryx Copilot retains message history for 90 days and encrypts Copilot data in transit and at rest. This section defines Copilot access scope and usage controls with Google Gemini.
Alteryx retains message history for a period of 90 days. For more information about security and data privacy with Alteryx, review the frequently asked questions for Alteryx Copilot , or go to the Alteryx Trust page.
How Alteryx Copilot Generates Responses with Google Gemini
Alteryx Copilot uses Google Gemini models hosted on Google Cloud’s Vertex AI platform to generate. Accordingly, Copilot’s security, privacy, and data-handling controls are based on those offered by Google Cloud.
How Alteryx Protects Your Data with Google Cloud Services
Alteryx configures Google Cloud services so that Copilot data is encrypted in transit and at rest using industry-standard encryption protocols. For authoritative and up-to-date details on Google Cloud’s security and data-protection practices, please refer to:
Google Cloud Security: https://cloud.google.com/security
Vertex AI Data Governance & Privacy: https://cloud.google.com/vertex-ai/docs/general/data-governance
Google Cloud Encryption Overview: https://cloud.google.com/security/encryption
What Data Copilot Can Access
By default, Alteryx Copilot only has access to workflow metadata, workflow XML, and your dataset metadata. Metadata is only sent to Alteryx Copilot when you send a new prompt. If you enable Data Awareness, Alteryx Copilot will also have access to the data in your workflow.
Model Training and Prompt Engineering
Alteryx Copilot doesn't train, retrain, or fine-tune AI models with your workflows or data. These points clarify how Alteryx Copilot uses AI models:
What Alteryx Copilot Does Not Do
Train or retrain foundation models (large language models).
Fine-tune models using your data.
Build new models from scratch.
What Alteryx Copilot Does Do
Use pre-trained third-party models.
Improve responses through prompt design and system instructions.
Supply structured context with each request.
Retrieve relevant product or workflow information at runtime to ground responses.
Evaluate and refine prompts and orchestration logic to improve accuracy, safety, and usefulness.
Why This Matters
Model training and fine-tuning require large datasets and permanently change model behavior. Prompt engineering and retrieval don't modify the model. They shape individual requests and responses.
From a compliance perspective, this distinction is critical. Alteryx Copilot behavior is driven by runtime instructions and context, not by retraining models on your data.