Privacy Policy

Last updated: 7 September 2026

What the product does

Therapeutical.ai is a VS Code extension that combines four-question check-ins, recommendations, AI token and cost tracking, an unread-code review queue, and a shareable work summary. It helps you decide whether to continue, clarify your next step, take a break, or stop.

This policy covers version 1.1.1 of the extension, its cloud features, and the Therapeutical.ai website. Work-state and technical estimates are experimental. They are not clinical assessments or proof of burnout, recovery, task success, or token savings.

Data on your device

Local tracking uses numeric activity signals from VS Code and, when enabled, connected AI tools. These include edit sizes and timing, session activity and inactivity, focus and navigation, Git activity, AI run outcomes, review coverage, and observed token and cost totals. Missing measurements remain unknown rather than being recorded as zero.

• Check-ins: four answers about mental drain, goal clarity, understanding AI changes, and frustration, plus a combined score, timing, questionnaire version, and scheduling context. Up to 180 records / 180 days. Incomplete attempts save no partial answers; prompt events record whether a check-in became eligible, appeared, was completed, dismissed, or expired, up to 500 events.
• Prediction examples: a numeric snapshot at each completed check-in, capture metadata, and later technical and work-state outcomes. Up to 180 records / 180 days. Incompatible examples from older versions are removed locally; original check-in answers remain subject to their own retention.
• Run evidence and review queue: up to 4,000 numeric run records / seven days, and review measurements for up to 200 hashed file identifiers / 14 days. File names are resolved in memory for the review queue and are not stored in this research history.
• Activity and summaries: daily totals for 30 days, up to 80 session summaries / 30 days, and one unfinished-session snapshot for restart recovery.
• Recommendation events and follow-ups: estimates, displayed suggestions, selected actions, and links to later observations. Events and follow-ups are each capped at 1,000 records / 180 days. They remain local.
• Notes and settings: a clarified goal stays in memory for the session. A resume note stays on your device until cleared or replaced. Pause settings, consent choices, scheduling state, anonymous access credentials, and upload queue records support the extension's operation. Authentication refresh credentials use VS Code SecretStorage.

Retention limits apply as the extension loads or updates its local storage. Exported files and images that you save are controlled by you and are not removed by these limits.

No baseline questionnaire is offered. Historical baseline answers are excluded from active uploads, prediction, and analysis; existing local and cloud records remain available for export and deletion.

How recommendation follow-ups work

A displayed recommendation and an action selection are recorded separately. Each follow-up links to the next two elapsed hours of observed editor activity, inactivity, run outcomes, and token coverage. It also links to the next completed check-in within seven days, preserving the actual time between the suggestion and the answer.

Tracking gaps, including time when VS Code is closed or suspended, are marked unknown. Clicking Break does not prove that a break happened, and editor inactivity does not prove rest or recovery. Follow-ups are observational records for future investigation; they are not inputs to the current prediction model and are not uploaded.

Cloud predictions and research sharing

Local check-ins, tracking, summaries, and review tools work without cloud participation. Cloud prediction requires your confirmation of the current disclosure and contribution of the described research data. The setup choice is initially selected, but upload starts only after confirmation. Predictions also need a recent completed check-in and enough usable history. Cloud features honor VS Code's telemetry setting.

For a prediction, the extension sends a bounded set of numeric examples, their outcome labels, and the current numeric snapshot through an authenticated service to an external hosted prediction processor. Inputs include technical activity and check-in values. The combined outcome uses both subsequent rework/review patterns and change in the four-answer work-state score. Missing or interrupted observations do not become favorable outcomes. They can include compatible examples collected locally before you enabled cloud participation.

Separately, with research sharing enabled, records created after your current consent time may be uploaded:

• Daily and AI-run technical measurements, outcomes, review coverage, and observed token/cost totals.
• Complete check-in answers and scores, questionnaire version, timestamps, scheduling context, and prompt events.
• Numeric prediction examples and their separately recorded technical and work-state outcomes.

These research records support longitudinal analysis and development of future Therapeutical.ai models. Recommendation events, follow-up links, local capture metadata, goal text, and resume notes are not added to these uploads.

An anonymous Firebase account authenticates requests. Research records use a coded participant identifier derived from that account, allowing records from one installation to be linked over time. This is pseudonymous data, not a guarantee that observations cannot be linked to a person.

Cloud research records receive a 365-day expiry when written or updated. Expiry deletion is asynchronous. Security logs, backups, and processing-provider records may follow separate retention schedules. The hosted prediction processor may temporarily retain submitted numeric inputs for processing, caching, or reproducibility; the research-store expiry is not a promise about those separate copies.

Content excluded from research and prediction

The extension does not upload source code, prompt text, AI conversations, file names or paths, hashed file identifiers, repository or branch names, commit messages, goal text, or resume notes in research or prediction requests. Email addresses, account names, and machine names are also excluded from those payloads.

Local editor and connector integrations may receive content or file paths as part of ordinary VS Code or AI-tool events. They reduce those events to numeric measurements or discard the content. Authentication credentials are sent separately to operate cloud access; they are not prediction inputs or research fields.

Your choices and data controls

The VS Code Command Palette and Therapeutical.ai: Data controls provide these options:

• Export local research saves check-ins, prompt events, prediction examples, recommendation events and follow-ups, plus any historical local baseline. It excludes your notes and authentication credentials.
• Delete local prediction history removes local examples, run evidence, recommendation events, and follow-ups. It keeps original check-in answers and cloud records.
• Delete local baseline is available when a historical baseline exists on the device.
• Show what would be uploaded previews queued research records. Stop cloud prediction & research stops new sharing and clears pending uploads; it does not delete records already uploaded.
• Export cloud data and Permanently delete cloud data operate on the current installation's anonymous account, including historical assessment records. Cloud deletion also removes that account and stops sharing. Local check-in answers remain on your device.

You can disable token tracking or disconnect an AI tool separately. Taking a break or stopping for the day pauses check-ins; local activity tracking can continue. Disable the extension in VS Code to stop new local collection.

For access, correction, deletion, withdrawal of research participation, or other privacy requests, contact terapeutiskiai@gmail.com. Depending on your location, you may also have rights to restrict or object to processing, obtain a portable copy, and complain to your local data protection authority.

Sharing your work summary

Summary images are generated on your device and include the activity totals visible in the share preview. They exclude check-in answers, private notes, goals, file names, recommendation follow-ups, and participant identifiers. Saving or copying a summary does not upload it to our research service.

Choosing a social-sharing option opens the selected service with summary text. You decide whether to publish. That service processes the information you send under its own privacy policy.

Services, website, and policy updates

We use Google Firebase for authentication, cloud functions, research storage, and website hosting, and an external hosted prediction processor for numeric predictions. Cloud infrastructure includes services in the United States; processing may occur outside your country. Service providers may receive routine connection and security information such as IP address, request time, and authentication metadata.

The landing page does not use advertising or analytics trackers. Website hosting still processes ordinary request and security information. If you contact us by email, your message and contact details are used to respond to the request.

We update this policy when the product or its data practices change. Material expansions of research uploads require an updated disclosure and confirmation before they begin.

Contact Therapeutical.ai: terapeutiskiai@gmail.com.