PreTensor Privacy Policy

Effective date: October 12, 2026

PreTensor processes Local AI, slide analysis, and local voice inputs on the user's own device. If the user explicitly selects PreDoctor API mode, the extension sends the submitted conversation request to the authenticated PreDoctor API on the same supported origin. It does not send data to advertisers, analytics providers, or unrelated origins.

Scope

This policy applies to the PreTensor Chrome extension and its local companion runtime. The extension's single purpose is to connect supported PreDoctor web pages to user-requested local NVIDIA CUDA processing and, only after explicit consent, to the authenticated PreDoctor conversation API on the same supported origin.

Data handled

When a supported PreDoctor page asks the user to run local inference, the extension may handle:

• user-provided prompts, chat messages, token identifiers, generation options, and inference results;
• medical text and images intentionally submitted for local inference, which may contain health information and personal identifiers such as a name, email address, patient identifier, or age;
• other website content intentionally passed by a supported PreDoctor page for local text or vision inference;
• local device and runtime diagnostics, such as GPU name, CUDA version, compute capability, memory availability, runtime readiness, and performance timings; and
• the selected local model identifier, used to restore the user's model choice in the extension popup; and
• a short-lived authenticated API credential when the user submits a request in PreDoctor API mode; and
• the address of the supported page solely to verify that the request originated from an allowed PreDoctor or local-development page and that an API request targets that exact origin.

The extension does not read or collect browsing history, cookies, passwords, financial or payment information, location, contacts, email inbox contents, user activity telemetry, or content from unrelated websites. It handles the short-lived PreDoctor API credential only for an active request and does not store it.

How data is used

For Local AI, slide analysis, and local voice, the extension forwards user-requested data through Chrome Native Messaging to the PreTensor companion runtime on the same computer. The runtime uses the data only to execute the requested local computation, return its result, report status, and diagnose errors. If the user selects and consents to PreDoctor API mode, the extension instead relays the submitted conversation request to the authenticated model-list or chat-completion endpoint on the exact current PreDoctor origin. The extension does not use data for advertising, profiling, credit decisions, or purposes unrelated to these requested features.

Network transfer and sharing

Local AI, slide analysis, and local voice traffic stays between Chrome and the local companion process. In PreDoctor API mode, the extension sends the submitted conversation, attached image or current-view context included by the supported page, and short-lived authorization credential only to the exact current PreDoctor origin. The relay accepts only the declared model-list and chat-completion paths, rejects redirects and unrelated origins, omits browser cookies, and bounds request and response sizes. If a user or supported page supplies an image URL for local processing, Chrome may retrieve that image from the URL's origin before processing. That origin's own privacy policy applies to the retrieval.

PreTensor does not sell user data. It does not share user data with advertisers, data brokers, or analytics providers. Disclosure may occur only when required by applicable law or when the user explicitly asks PreDoctor support to inspect specific diagnostic information.

Local storage and retention

The extension itself does not persist prompts, images, inference results, API credentials, or browsing activity. It stores only the selected local model identifier in Chrome's extension-local storage so that the same model remains selected after the popup closes, plus a short-lived privacy-bounded runtime-update observation in browser-session storage during host replacement. The companion runtime may retain model files, model download and verification state, runtime configuration, compiled GPU assets, and local cache data needed for performance. Active inference buffers and KV cache data are kept locally and are released when sessions end or the runtime stops. Users can clear the model choice by removing the extension and can remove the local runtime and its downloaded model files with the PreTensor uninstaller.

Permissions

• nativeMessaging connects the extension to the local PreTensor companion runtime.
• storage stores only the selected local model identifier in extension-local storage.
• Access to https://predoctor.ai/* exposes the bridge only on the production PreDoctor Viewer origin and permits the explicitly selected same-origin PreDoctor conversation API request.

Security and Limited Use

PreTensor limits access to the minimum permissions required for its single purpose and verifies the extension origin before the companion runtime accepts a request. The use of information received from Chrome APIs adheres to the Chrome Web Store User Data Policy, including the Limited Use requirements.

Changes

If PreTensor's data practices change, this policy and the Chrome Web Store disclosures will be updated before the changed behavior is released. Material changes will also be disclosed in the product interface when required.

Contact

Questions or deletion requests concerning locally stored PreTensor data can be sent to support@predoctor.ai.