Effective date: 1 July 2026
1. What Zero Data Retention (ZDR) means
Zero Data Retention means that after your API request is processed, its content is not kept. The prompt goes in, the answer comes back, and the request is discarded. ZDR is a retention guarantee: it does not mean the data never moves, and it does not cover data you deliberately store.
2. Our default posture
Perceptron is transient by default. For stateless inference calls we do not persist the body of your request or the model’s response after delivery. We retain only the metadata needed to bill and secure the request, such as the model used, token counts, latency, and cost. You do not need to enable a setting to get this, as it is the default.
3. What ZDR covers
Chat and other text inference through the API
Speech synthesis and transcription requests
Vision requests
Web search queries
Any other stateless endpoint that processes a request and returns a result.
For these, request and response content is processed transiently and not retained after the response is delivered.
4. What ZDR does not cover
Some features exist precisely to store things. These persist until you delete them, and are outside ZDR:
- Persona conversation history, tasks, and memory.
- Persona workspace and filesystem contents, and uploaded files.
- Perceptron Studio video, image, animation, and sound artifacts.
- Feeds subscriptions and domain name configuration.
- Box records and screenshots.
- Billing, ledger, and mining reward records.
- Bounded operational logs, tracing, and security records.
ZDR applies to processing, not to storage you request. If you want the strongest privacy posture, use stateless inference and avoid features that persist data as listed above.
5. Caching
Perceptron may briefly cache a prompt in memory to speed up a response. We treat that transient caching as processing rather than retention: it is short-lived, exists to serve the request, and is not kept for later use.
6. Mining
If you route a request to miners, your prompt is processed on another user’s hardware, and Perceptron cannot control what that device logs locally. When a request is served by your own minig nodes, or by our own infrastructure, the default transient handling applies. If ZDR is essential for a request, do not enable miners. Miners are diabled by default for API requests.
7. No training on your content
We DO NOT train models on your content, and we DO NOT use it to improve our own models.
8. Limits
ZDR does not prevent us from retaining metadata needed for billing, fraud prevention, security, or legal compliance, and it does not apply to data you choose to store. We may retain information where the law requires it.
10. Contact
Questions about ZDR: mail@bartmoss.org