Security
Controls that are enforced in the product
This page is maintained by THE SHRIKS to answer common security questions about LokiAI. It describes controls that are implemented today, not certifications.
- Local inference
- Device-scoped credentials
- Verified artifacts
01Access and identity
Strict Security Boundaries
Control Plane
- [Encrypted Provider Keys]
- [Tenant Isolation]
- [Rate Limits]
← HTTPS / WSS →STRICT CORS
Edge Device
- [Device-Scoped Creds]
- [Local Execution]
- [Offline Inference]
02Data handling
Where data goes
- Inference inputs — prompts, frames and audio — are processed on the device and are not routed through LokiAI infrastructure.
- Model weights are downloaded by the device directly from the source CDN; the control plane does not proxy them.
- Provider keys are stored encrypted at rest and are never returned to the browser in raw form.
- Deployment records store artifact metadata and state transitions, not payload content.
03Shared responsibility
Where our boundary ends
LokiAI provides
Isolation between accounts, credential issue and revocation, artifact verification and transport security.
You are responsible for
Physical control of paired devices, who you grant account access to, and the models you choose to deploy.
Reporting
If you believe you have found a vulnerability, contact us before disclosing it publicly.
Security questions before a deployment?
Ask us directly — we would rather answer specifics than publish absolutes.
