Data processing by third-party AI services

Data used to produce a response

When you submit a cloud request, the selected service may receive your message, necessary conversation history and instructions, attached images and files or text extracted from them, tool queries and tool results. Technical request information is also processed for delivery, usage accounting and security. The exact data depends on the feature and selected model.

Messages and files may contain health or financial information if you choose to add it, including in conversations with informational personas. You do not need to provide medical records, card details or identifying financial documents to use the app. Remove names, document numbers and other unnecessary information; do not share another person’s data without a lawful basis.

Services handling model requests

Empatra Studio’s cloud infrastructure uses Alibaba Cloud, UniAPI and CometAPI. The recipient of an individual request depends on the selected model and its processing route; a request is not necessarily sent to every listed service.

Alibaba Cloud

Alibaba Cloud provides cloud infrastructure and model APIs, including Model Studio. Processing depends on the product, region and workspace configuration, and is subject to Alibaba Cloud’s terms.

Alibaba Cloud privacy policy ↗

UniAPI

UniAPI is an API service for accessing models. It processes and routes requests to model providers, whose own policies may apply at the next stage. The link leads to its official website; the link itself does not establish contractual data-processing terms.

UniAPI official website ↗

CometAPI

CometAPI is an API service for accessing models. It receives requests and provides processing through available models and connections, subject to its published terms and the rules of participating providers.

CometAPI privacy policy ↗

A model’s name is not always the name of the service receiving the request. An API intermediary may forward necessary data to the next model provider. Processing participants and locations depend on the route, and processing may occur outside Russia. Consider the external services’ terms before sharing confidential information.

Screening material before publication

Versions with publication screening enabled separately screen the first version shared by link and each public update. This uses the title, format and necessary content of the specific version you explicitly submit for publication, including supported images. Saving private material alone does not trigger this transmission. Other private conversations, access keys and closed agent threads on your computer are excluded.

Before sending data to external AI, the app shows the recipients, specific purposes, data and current policy. One explicit permission may cover the listed cloud AI features, including screening of materials you deliberately submit for publication. If valid permission covers that purpose and recipient, publishing requires no additional permission prompt. Accepting the offer or earlier permission limited to cloud chat is not automatically expanded. The operator configures the screening recipient; a model name does not identify it, and material is not sent to every service listed above. Declining preserves basic access and private materials, but a new public version may remain unavailable. Closing a link requires no new consent.

Provider failure, an insufficient budget, an uncertain answer or an unsupported visual or active element does not authorise new publication. You may edit the material or request manual reconsideration through support. While an update is screened, the previously approved public version may remain accessible while the new version stays private. Manual approval permits a publication retry and does not open a link on the owner’s behalf. An AI result alone does not cause an account ban or payment withholding. Reporting and subsequent moderation of already shared materials remain available.

Before enabling screening, the operator discloses the actual recipient’s terms, processing location and retention periods for submitted materials and decision records. Screening does not authorise indefinite additional copies or use for advertising or training. Reviewed-version, decision-record and provider retention periods may differ; zero retention must not be assumed. Withdrawing permission stops future transmissions for this purpose but does not reverse completed requests. For retention information or deletion requests, contact the address below.

Separately: web search and page reading

When web search or page-reading tools are used, Serper may receive the search query, search parameters or the URL of a page to retrieve. These may include part of your question. Serper is used for search and page retrieval, rather than receiving every ordinary message. Retrieved information may then become part of a model request’s context.

Retention and provider terms

External services have different retention, troubleshooting, abuse prevention and onward processing rules. Empatra Studio does not promise no logging, zero retention or no model training for every service and route. Current provider terms are linked on this page.

Alibaba Cloud Model Studio’s published standard retention policy keeps prompts and responses for up to 30 days. Safety-flagged data may be kept for up to two years, and law may require a different period. Zero Data Retention (ZDR) is available for certain enterprise workspaces and supported models only when enabled and its conditions are met. Its availability does not mean Empatra requests use ZDR. Model Studio terms must not automatically be applied to other Alibaba Cloud products, including Content Moderation.

A model name or marketing description does not establish UniAPI retention, training, legal entity, onward processors or processing region. The operator must confirm the terms of the specific connection before using it for publication screening. A screening route is not enabled if necessary facts have not been confirmed and disclosed.

Computer control and cloud AI use different paths

A remote controller connects to your Empatra Runtime using end-to-end encryption through the Blind Remote Relay: the intermediary forwards encrypted commands and state. This does not mean request data remains encrypted from the selected AI provider. When the agent on your computer calls an external model or tool, that service processes the necessary content according to the host configuration and provider terms.

Your choices and requests

Before the first cloud transmission, the app shows specific recipients, data and processing purposes. One explicit permission may cover the listed features: chat, images, search and screening of materials you deliberately submit for publication. It is separate from accepting the offer and is remembered: individual requests and publications require no new confirmation within the disclosed terms. Declining preserves your account, basic access and control of your own computer; external requests made by that computer depend on its agent settings. Earlier permissions retain only their originally disclosed scope.

You can withdraw permission in privacy settings; available controls depend on your installed app version. Withdrawal applies to future requests within the withdrawn scope, does not reverse completed processing and does not replace a separate data-deletion request.

For questions about recipients, processing, data access or deletion, contact the Empatra Studio operator, individual entrepreneur Roman Viktorovich Sedykh, at the address below. Do not include secrets or sensitive documents.