Data Privacy & Governance at ModelStream

At ModelStream, your data sovereignty is our baseline. Whether you are interacting through our professional workspace or integrating via our high-performance API, we prioritize a transparent data lifecycle where you maintain absolute control over your inputs and outputs.

This overview details how your information is managed within the ModelStream ecosystem. For a deeper legal dive, please refer to our Privacy Policy and Terms of Service.

Core Data Handling: Content Sovereignty

Data Collection Practices

Last Updated: 21 April 2026

At ModelStream, our core philosophy is zero-retention passthrough. We do not store, log, or persist the contents of your inference requests (prompts, images, etc.) or the generated responses (outputs).

We collect only non-content metadata strictly necessary for routing, billing, and system observability.

What we do not collect:

• Prompt text, system instructions, or system messages
• Input images, audio, or video files
• Model-generated text, image, audio, or video outputs
• Fine-tuning datasets or evaluation datasets

What we do collect (Metadata):

• Routing metadata: Target AI Model, region, latency, HTTP status code
• Billing metadata: Token counts (input/output), image dimensions (if applicable), duration
• Observability metadata: Timestamp, Workspace ID, IP Address (retained for 30 days for security analysis)

This metadata is never used to train generative models.