Hi all,
Posting here per Q11 of the FAQ, after going through the relevant community resources first — the API FAQ, Denzel-Warren's November 2024 policy posts on AI and data display, and several recent threads from developers in similar situations (the Athlete Limit Increase megathread, ProPace, AthletIQ AI, Rule9, and the Kukiman0/dlandre thread on AI inference). Status below, AI specifics framed against the November 20 clarification, and a clear ask at the end.
— Submission status
- Client ID: 225209 (Zenith)
- Submitted: 16 April 2026 (31 days ago)
- Follow-ups to developers@strava.com: 4, 5, and 10 May — no reply on any
- App page updated today (17 May): Terms of Service URL added, website moved to HTTPS, description rewritten to detail the AI use case
— Use case
Personalized coaching for serious amateur endurance athletes (running, cycling, triathlon). Athletes record short voice notes after each session capturing how the workout felt. Strava activity data is paired with these qualitative notes per athlete, and the coach answers conversational questions grounded in that athlete's own history.
Per Denzel-Warren's November 20, 2024 clarification, Zenith is explicitly within the allowed category: "coaching platforms focused on providing feedback to users and tools that help users understand their data and performance." It is a feedback tool for the individual athlete, not a training-data source.
— Where Strava data appears in the product
- Memory / Day detail — daily timeline pairing activities with voice notes; includes "View on Strava" link
- Profile — data sources view with "Powered by Strava" attribution
Strava data is also consumed in-memory by the coach to ground answers, but not displayed as raw Strava content outside the two surfaces above. Annotated screenshots available by email on request.
— Scale
Private beta, capped at 50 active accounts. 12-month projection: 200–500 connected athletes.
— How AI interacts with Strava data — inference only, no training
LLM provider: Google Gemini, called per-request only.
Per-request inputs:
- The athlete's own voice notes (transcribed) + conversation history
- A slice of the athlete's own activity data (last N activities or those they reference)
- Retrieved embeddings of the athlete's own past notes (RAG)
Per-request output: a natural-language response returned only to that athlete.
Explicit guarantees (also in the public Privacy Policy):
- Strava data is never used to train, fine-tune, or develop any model. Per Google Cloud DPF terms, Gemini API inputs are not used for model training
- Strava data never leaves the request/response cycle for any cross-user purpose
- No third-party sharing, no cross-user analytics, no derivative product
- Per-user data isolation enforced at the DB layer (Row-Level Security on Supabase)
— Compliance
Confirmed compliance with the API Agreement and Brand Guidelines. "Powered by Strava" attribution shown where required. No virtual races, no replication of Strava sites/services.
— References
- Terms: https://coach-zenith.com/terms
- Privacy: https://coach-zenith.com/privacy
- Sub-processors: https://coach-zenith.com/subprocessors
— What I'm asking
An athlete cap increase to 50 connected athletes to run the private beta. The application has been complete since April 16 (31 days), the AI specifics required by Q3 of the FAQ are addressed above, and the use case fits the November 20 clarification. If something is missing on my side, point it out and I'll address it.
Thanks,
Jordi
jordi@coach-zenith.com