Global Privacy Data Policy

When you provision a new cluster on our platform, you might be required to submit your corporate credentials (name, enterprise domain, billing address, contact number). Please understand that you have the capability to browse our documentation anonymously.

There exist three primary methods by which enterprise / tenant telemetry is ingested into our systems:

• Data provided by administrators for executing a compute benchmark through the Infrastructure Evaluation portal.
• Metrics transmitted by workloads for scaling a model deployment through the Resource Auto-scaler API webhook.
• Credentials submitted by engineers for joining the Tensorik Developer Beta Program.
• Analytics captured via tokens for the optimization of API latency.

We also monitor diagnostic metadata such as subnets, IP routes, type of virtualization stack, and/or architecture from cluster endpoints as part of maintaining the reliability of this grid. This telemetry is aggregated securely. By connecting your nodes without disabling tracking protocols, clients grant explicit consent to allow Tensorik to process this diagnostic information.

Tensorik shall analyze operational metadata ingested through endpoints to orchestrate distributed workloads and respond to system bottlenecks by allocating resources to containerized services and databases that will sustain your global infrastructure.

Any encryption key submitted on our portal is considered sensitive and only accessible to verified microservices which are configured to keep the certificates encrypted. Cluster configuration files are processed by our automated deployment pipeline engine and utilized solely for the environment for which the provisioning was executed.

Models uploaded by developers while training AI models through our platform will remain proprietary. It may be cached by our internal hypervisors for the purpose of current and future processing with us.

Logs generated when integrating with the Enterprise Partner API is used for identifying the authenticated service and rate limiting. We may also send them security alerts in addition to latency reports.

Telemetry collected autonomously is used to help us scale clusters faster. This data enables us to understand what microservices utilize RAM, how often they peak, and what zones of the network they query most heavily. We use this information to optimize our edge-based routing and also to defend our architecture from DDoS attacks.

We take workload isolation seriously and deploy rigorous cryptographic protocols to safeguard your proprietary neural weights. We do not expose or leak your computationally generated models to third parties. We may share telemetry with trusted auditors who assist us in certifying our datacenters, optimizing hardware, or scaling infrastructure, provided that those auditors agree to keep this telemetry classified.

You retain the right to terminate your cluster and initiate the purging of your models from our servers. If you wish to opt-out of our telemetry or want your endpoints detached, please command our API layer. We will process your execution promptly in alignment with stringent hardware sanitation protocols.

Our clusters are not configured for access by unauthenticated systems outside your VPC firewall. We do not intentionally process rogue traffic from scanners. If you are a DevSecOps engineer and discover your external subnet has forwarded us with unexpected traffic, please alert us immediately, and we will apply patches to securely drop such packets.

Our ecosystem may maintain hooks to external services. Please be advised that we are not liable for the uptime or security posture of such external endpoints. We instruct our developers to be cautious when they proxy our mesh and to audit the security certificates of any external API that ingests sensitive enterprise payloads.

Nodes on this grid utilize “tokens”, which are encrypted signatures that the broker issues to your containers or edge devices to make your handshakes more performant. These keys are used for authorization the next time you ping us. Tokens help us maintain the throughput of the pipeline and allow us to remember your configurations and deploy scaling based on query frequencies.

We additionally use tokens to monitor your operations across our instances, such as the nodes you spawn and the bandwidth you consume. This telemetry helps us track resource utilization across the cluster.

Certain of the tokens and agents are injected here by integrated observability partners whose daemonsets we have embedded into the kernel, and these can similarly influence your tracing of other clusters you might debug after exiting ours. Such agents do not retain any confidential weights but are designed for uniquely fingerprinting your workload and virtual machine. If you do not authorize these agents, you will receive less granular observability.

We likewise export metadata regarding your utilization of our stack with our security posture, compliance, and auditing vendors who may correlate it with extra telemetry that you've submitted to them or that they've aggregated from your adoption of their toolchains.

We mandate critical tokens for the authentication of our backends. Furthermore, we may deploy diagnostic tokens to analyze how you interface with our APIs, and logging tokens to stream relevant alerts to you based on your thresholds.

You can rotate and invalidate tokens via multiple endpoints. Please keep in mind that revoking or expiring tokens will instantly terminate your session state and sections of our dashboard may no longer be dynamically rendered. Most SDKs automatically cache tokens, but you can dictate whether or not to store tokens via your environment variables.

By calling our endpoints, you agree to our security framework. All modifications to our governance model will be committed to this repository. We advise you to pull this documentation regularly to remain updated regarding how we are securing your infrastructure.