Privacy Policy
Last updated: October 12, 2026
This policy describes how Exla ("Exla", "we", "us") handles data when you use Redline, our inference API and website at redline.invariantmachines.com (the "Service").
Prompts and completions
- We do not retain prompts or completions. Request and response content is processed in memory to generate a response and is not written to persistent storage or logs.
- We do not train on your data. Prompts and completions are never used to train or fine-tune models.
- We do not publish or sell your data.
- Prompt caching keeps derived model state (KV cache) for repeated prompt prefixes in volatile memory for a short period (typically minutes) solely to serve your subsequent requests faster. It is not human-readable, is not shared across customers, and is discarded automatically.
Operational metadata
We collect request metadata needed to operate and bill the Service: timestamps, model, token counts, latency, throughput, error codes, account or API-key identifiers, and the IP address of the caller. We use it for billing, capacity planning, abuse prevention and reliability. Metadata is retained for up to 90 days, or longer where required for billing records.
Website
The website does not use advertising trackers. Our hosting provider may log standard web-server data (IP address, user agent) for security.
Sub-processors
We use infrastructure providers for compute, networking and hosting. When you access Redline through OpenRouter, OpenRouter's own privacy policy also applies to data you send to OpenRouter.
Security
Traffic is encrypted in transit with TLS. Access to production systems is restricted to authorized personnel.
Contact
Questions or requests: hello@invariantmachines.com.