Pular para o conteúdo
AvraAvra
Esc
↑↓navegar↵abrir⌘Jpré-visualizar
Nesta página

Data privacy & compliance

Who owns your data, what Avra's models learn from it, where it flows in Avra's cloud and in your environment, and how Avra handles LGPD.

We don’t share information between customers.

Your data is isolated by tenant and workspace, in the environment you choose. Models specific to you are exclusive to your workspace, and your raw data is never exposed to or used by other customers.

This page covers who owns your data, what Avra’s models learn from it and what they don’t, where it flows when Avra runs in its cloud or in your environment, and how Avra handles LGPD, retention, and deletion.

Data ownership

Any data you provide — CRM records, transactions, outcomes — remains your exclusive property. Avra acts as a data processor on your behalf.

The models specific to you — adapted and fine-tuned copies of Avra RFM, and downstream models trained on your outcomes — are exclusive to your workspace and never used to serve another customer. They’re derived from Avra’s models and licensed to you for use in your workspace.

As for Avra’s shared models, the RFM is pre-trained on diverse relational data, and Avra GFM on public, private, and licensed sources. No customer data goes into either.

This is set out in your Master Service Agreement.

What models learn from your data

Your data meets Avra’s models in three ways, shown below with what changes in each.

No arrow leads back to the shared models: no customer data goes into them, and copies are never merged back.

How your data is used What changes
In-context prediction Your examples inform each prediction, and no model weights change. Prepared data and caches can persist, inside your isolated environment
Adaptation and fine-tuning A copy of the RFM, exclusive to your workspace. It’s a one-way fork: never merged back into the shared model
Avra GFM downstream models A separate model trained on GFM outputs and your data. The GFM itself isn’t trained

What Avra learns

Avra learns which architectures, training strategies, and data structures work across use cases. That’s engineering knowledge, and it improves Avra’s models. It isn’t permission to pool customer data, share customer models, or merge customer-specific weights into a shared model.

Data isolation

Your data and the models specific to you stay in your workspace

Your data, the models specific to you, and the predictions they make are never accessible to other customers.

Avra enforces that through tenant isolation and access control. Each customer’s data and models are siloed by tenant and workspace, and access to customer data is role-based and granted on a need-to-know basis. See the figure in Security overview for how the boundaries nest.

When Avra runs in its cloud, your data takes this path. Every delivery is encrypted in transit and validated against your agreed data contract before a model can use it, and a delivery that fails is rejected, not published. What passes is encrypted at rest in your tenant and workspace.

Running in your environment

For regulatory, sovereignty, or sensitivity requirements, Avra can run inside your own environment. Your data stays in your environment, with Avra’s data layer running next to it. Avra RFM runs in your AWS account as an AWS SageMaker algorithm, under license. Its jobs run network-isolated, and its weights and code are never delivered as files. Adaptation and fine-tuning artifacts are written to your storage, encrypted.

Your environmentYour datastays in your environmentAvra’s data layerruns next to your dataYour AWS accountAvra RFMSageMaker algorithm,under licensenetwork-isolated jobsYour storageadaptation andfine-tuning artifacts,encryptedAvraControl planedashboard, API,model registryAvra GFMAvra’s cloud onlymetadata,telemetryoutputs
When Avra runs in your environment, your data, Avra’s data layer, and Avra RFM’s jobs are on your side. What returns to Avra is execution metadata and operational telemetry, labeled with your cluster name. The control plane stays with Avra, and Avra GFM runs only in Avra’s cloud, sending you its outputs.

From your environment, what returns to Avra is execution metadata, such as each command’s status and result; operational telemetry labeled with your cluster name, which covers logs from Avra’s pods, the metrics and traces of Avra’s components, and resource metrics for the collector’s node and the pods on it; and the deployment agent’s release status and logs. See the data layer’s security details. Your data, the models adapted to it, and their outputs stay in your environment. The control plane, meaning the dashboard, the API, and the model registry, stays with Avra. Avra GFM runs only on Avra’s cloud, and you receive its outputs.

Your agreement sets out the license terms for running Avra in your environment. See Deployment for how it works.

Regulatory compliance

Avra’s processing is LGPD-compliant. Data protection terms are set out in your agreement.

Data retention and deletion

Retention and deletion follow your agreement.

For privacy questions, see Avra’s privacy policy or contact your Avra representative.

Where to go next

Esta página foi útil?