Data Policy
This Data Policy describes exactly how Consensus Protocol LLC handles data submitted to its AI inference API. It supplements our Privacy Policy and is written for users and platforms that need a precise statement of our logging and retention practices.
1. Summary
Zero retention of inference content. Prompts and completions are processed in memory on our own hardware and are never written to disk, never logged, never used for training, and never reviewed by humans. Only operational metadata is retained, and only for up to 90 days.
2. Prompts and Completions
- Storage: None. Request and response content exists in system and GPU memory only for the duration of processing and is discarded when the response is delivered.
- Logging: Application and access logs do not contain prompt or completion content.
- Training: We do not train, fine tune, or evaluate models on inference content.
- Human review: No employee or contractor of Consensus Protocol reads inference content.
- Caching: Where prompt caching is enabled to improve performance, cached key value states are held in volatile memory on the serving hardware, are not inspectable as text, and expire automatically. They are never persisted to disk.
3. Metadata We Retain
For billing, settlement, capacity planning, and abuse prevention, we retain the following per request metadata for up to 90 days:
- Timestamp and request identifier
- Model identifier
- Input and output token counts
- Latency and throughput measurements
- HTTP status code and error class
After 90 days this metadata is deleted or aggregated into anonymous statistics that cannot be linked to any request.
4. Infrastructure
Inference runs on dedicated GPU hardware owned and operated by Consensus Protocol. We do not subprocess inference to third party compute providers. Network traffic to our API transits Cloudflare for routing and DDoS protection; Cloudflare does not receive persistent copies of inference content.
5. Encryption
All API traffic is encrypted in transit with TLS 1.2 or higher. There is no inference content at rest to encrypt, because inference content is not stored.
6. Access Through Routing Platforms
When you reach our API through a routing platform such as OpenRouter, the platform handles your account, authentication, and billing under its own policies. We receive requests without your identity attached. Any logging performed by the routing platform is governed by that platform's own data policy, not this one.
7. Legal Process
Because inference content is not retained, we are unable to produce past prompts or completions in response to legal process. Retained metadata may be disclosed if we are legally required to do so.
8. Changes
If our practices ever change in a way that reduces the protections described here, we will update this page and its effective date before the change takes effect.
9. Contact
Questions about this policy can be sent through the contact form at consensusprotocol.org.