Subprocessors.

TrackForge uses a small number of third-party services to run the product. Where one of them processes personal data on our behalf, it is a subprocessor and it is listed here.

The list below identifies each subprocessor, the data it receives, and whether it is active.

This page sits alongside the Privacy Policy, which covers everything else we do with personal information.

1. Who processes data for us

Eleven services, each with what it does for TrackForge and what it receives. The status word after each name is defined in section 2.

• Convex Live

  Application backend and database. Stores substantially all TrackForge data.

  What it receives. Everything the product stores: profiles, training plans, workouts and completions, personal records, injuries, wellness check-ins, performance observations, uploaded media, consent records.

• Netlify Live

  Hosts the TrackForge website and application. Netlify Forms processes the public waitlist form on TrackForge’s behalf.

  What it receives. Request IP address, browser user-agent, requested page, referrer, response status, country, and the waitlist email address you choose to submit.

• Zoho Live

  Runs the mailboxes behind the TrackForge addresses we publish, so mail you send us arrives somewhere. This is the receiving side only. Mail that TrackForge sends to you goes out through Resend instead. Zoho receives nothing from the application itself and holds no athlete records.

  What it receives. Only what you choose to send us: your email address and the contents of your message, when you write to one of our published addresses.

• Clerk Live

  Authentication, session management, and email verification. Holds account identity credentials.

  What it receives. Name, email address, authentication credentials, session and device metadata.

• Anthropic Live

  AI model provider for coaching, planning, technique analysis, Knowledge drafting, Research evaluation, and Research discovery.

  What it receives. Training-plan and coach-chat context; absence-note text; spreadsheet mapping requests; redacted coach material and cross-season facts used to draft inactive Knowledge proposals; Research resource text used for evaluation and summarisation; Research discovery topics, target URLs, and page content fetched through Anthropic web tools; and, for technique analysis, the event and phase labels plus athlete note supplied with the clip, extracted frames, on-device pose angles, gait events, and the stride contact, flight, and step timing timeline included in the prompt.

• OpenAI Inactive

  AI client code exists, but no current product path selects OpenAI.

  What it receives. Nothing while no product resolver selects it.

• Google Conditional

  Gemini-grounded discovery in Research, active only where a Gemini key is configured.

  What it receives. The discovery text a user enters. It does not receive technique frames or athlete training records through this path.

• Resend Conditional

  Transactional email: invitations, notifications, billing notices.

  What it receives. Recipient email address and the content of the message being sent.

• Sentry Conditional

  Crash, error, and application-performance monitoring.

  What it receives. Error and crash diagnostics plus sampled navigation and request timings, including the page or route and browser and device type. TrackForge is not configured to attach your IP address or account identity, and it does not send training, wellness, or chat content.

• Firecrawl Conditional

  Live web search behind the Research Library’s Discover feature.

  What it receives. The search text a user enters in Discover, and the page addresses it is asked to fetch. It does not receive athlete records.

• Stripe Not yet live

  Payments and subscription billing.

  What it receives. Billing contact details and payment information. Card details go to Stripe directly and never reach TrackForge systems.

2. What the status words mean

• Live. Processing personal data today.
• Conditional. Wired into the product, and active only where its configuration is present on a given deployment.
• Not yet live. Integrated, but the feature it serves has not launched. No payment has been processed through TrackForge.
• Inactive. Client code exists, but no product path currently selects the provider.

3. Three notes on the list above

On OpenAI. No current product resolver selects OpenAI, so it receives no data. It remains listed because client code exists; this status changes before any caller ships.

On Google. Google is reachable only for Gemini-grounded Research discovery where a Gemini key is configured. That path receives the discovery text entered by the user, not technique frames or athlete training records.

On AI training. We do not permit any of these companies to train models on your data. Anthropic’s commercial terms state that content submitted through their API is not used to train their models, and OpenAI’s API terms say the same for content submitted through their API. Google is not configured on any deployment today; if it is ever switched back on, it will be on a tier whose terms carry that same commitment. We operate no model of our own, so there is nothing for us to train.

4. Processing that does not use subprocessors

Pose estimation itself runs on the user’s device.

On-device pose estimation. Technique analysis measures body position using a model served by TrackForge that runs inside your browser. That pose-estimation step sends no frame or body measurement to another company. Separately, extracted frames are sent to Anthropic for the technique analysis disclosed above, and pose measurements reach TrackForge with the submission you choose to upload.

5. What none of them do

• We do not sell personal data. Not to any of these companies, not to anyone else.
• We run no advertising. No advertising networks, no ad trackers.
• We use no third-party product analytics and no cross-site tracking. Where Sentry is configured, it receives a 10% sample of application navigation and request timings for operational performance monitoring, as disclosed above; it receives no training content. Product-usage records stay on TrackForge servers and are not sent to an analytics vendor. See the Privacy Policy for what those records contain and how to switch them off.
• No other company trains its models on your data. The AI providers above process your prompts to answer them and are contractually barred from training on them. TrackForge itself may use service data to improve TrackForge, including training its own models, on the terms set out in the Privacy Policy: forward-looking only from August 17, 2026, never on injury, wellness, technique footage, body measurements or under-18 data, only with an institution’s written authorisation where the data came through one, and always subject to an individual opt-out.

6. Changes to this list

We will publish a change here before a new subprocessor begins processing personal data. Customers under a Data Processing Agreement receive at least 30 days’ notice and may object, on the terms set out in that agreement.

The date at the top of this page is the last time it changed.

7. Questions

Write to us at privacy@trackforge.fit, or see our Privacy Policy. To report a security vulnerability, see our security policy.