Skip to main content

AI inference with Strava data — is it prohibited under the new API Agreement?

  • May 7, 2026
  • 7 replies
  • 460 views
Forum|alt.badge.img+1

Hey everyone,

I'm a developer planning to build a cycling and running training app integrated with the Strava API, and before I start I'm trying to get clarity on the AI restrictions introduced in the November 2024 API Agreement update.

I understand that using Strava data to train or fine-tune AI models is clearly prohibited. What I'm less sure about is whether the restriction also covers inference-only use cases, specifically:

1. Sending a user's own Strava data as context in a single prompt to an external LLM API (Claude, GPT-4, Gemini etc.) to generate a personalized training plan or coaching feedback — no model training, data used only for that one request, result shown only to that user

2. Running a self-hosted local model on my own infrastructure (e.g. LLaMA, Mistral), where Strava data never leaves my server and zero training on Strava data occurs

Has anyone dealt with this and got a clear answer from Strava — either via the review process or by emailing developers@strava.com?

I've already sent an email to Strava asking for clarification, just curious if anyone here has been through this already.

Thanks!

    7 replies

    dlandre
    Forum|alt.badge.img+4
    • dlandre
    • Hub Rookie
    • May 10, 2026

    My assumption is that I need to firewall Strava data from any AI interaction, either on external infrastructure or my own.  Right now I have a clear firewall in code between any Strava data and any AI functionality.  Sadly.  If you find out differently that would be very interesting.  

      Forum|alt.badge.img
      • mauriblint
      • Hub Starter
      • May 21, 2026

      Hey ​@Kukiman0, do you have any news or more info? I want to start using the Strava API but not sure if they allow this kind of use cases.

      Thanks,

      Mauricio

      Forum|alt.badge.img+1
      • Kukiman0
      • Author
      • Hub Rookie
      • May 27, 2026

      I mailed them about 3 weeks ago, didnt receive any information so far

      Forum|alt.badge.img
      • gogrim
      • Hub Starter
      • June 15, 2026

      Still no answers from them? ☹️

      Forum|alt.badge.img
      • Adrabins
      • Hub Starter
      • September 11, 2026
      Following up on this thread with a specific case — I emailed developers@strava.com
      about this exact question and got an automated reply saying capacity/API questions
      aren't reviewed via email and to ask here instead.
       
      App: TyRun (single-player/testing mode, default athlete capacity).
       
      Use case: an athlete connects their own Strava account (read-only). Our backend
      imports that athlete's own activity data from the Strava API. Derived numeric
      metrics (distance, pace, HR, elevation — no name, no Strava ID, no raw GPS) from
      that athlete's own data are sent per-request to a third-party LLM purely for
      inference (no training/fine-tuning of any model, no retention beyond the single
      request). The output (a training plan or session analysis) is shown only to that
      same athlete.
       
      Does Section 5.3 ("development, training, evaluation, or operation of any AI
      Application") prohibit this — single-athlete, inference-only, first-party,
      display-only-to-the-data-owner — or is it aimed specifically at using Strava
      Data to build/train/improve AI models (including for third parties)?
       
      Would really appreciate an official answer, even just a yes/no on this narrow
      case — happy to adjust our implementation based on it.
      Jan_Mantau
      Forum|alt.badge.img+26
      • Jan_Mantau
      • Hub Powerhouse
      • September 11, 2026

      There are no official answers in this forum. But you should assume that any mention of AI leads to a rejection.

      Primed
      Forum|alt.badge.img
      • Primed
      • Hub Starter
      • September 15, 2026

      Adding another data point here, as I asked Strava essentially this same question on 7 May.

      I’m the founder of Primed, an AI-native endurance coaching platform. Before launch, I contacted Developer Support specifically to ask whether an athlete could authorise us to use their own Strava data for per-user LLM inference, where there is no model training or fine-tuning and the resulting coaching is returned only to that athlete.

      Ticket #14583 did not answer the question. Instead, Strava Developer Support directed me to this Community Hub for general API questions.

      More than four months later, I still haven’t been able to find an authoritative Strava answer here, while several other developers are now asking substantially the same question.

      The June policy wording appears much clearer: Section 5.3 seems to prohibit even inference-only use, including derived Strava data being placed into an AI context window or working memory.

      There have been a number of interpretations and assumptions in the Community about what Strava will or won’t approve. Those may be well intentioned, but assumptions are not enough when developers are making product, engineering and compliance decisions based on Strava’s rules.

      What is needed is an authoritative answer from Strava.

      If anyone from Strava is reading this, I’d particularly appreciate an official answer to two questions:

      1. Is athlete-authorised, inference-only use of Strava data by an independent AI product deliberately prohibited, even where there is no model training and the result is provided solely to that athlete?

      2. If so, is there any Extended Access, partnership, review or other pathway through which a legitimate AI-native fitness product can obtain approval?

      This is becoming increasingly important as Strava itself now provides an authorised route for athletes to use AI with their training history, while independent developers appear unable to establish whether comparable athlete-authorised use is permitted.

      I’ve followed up directly with Strava as well. However, given that Developer Support originally directed me to this Community Hub, I think it would be valuable for Strava to document its position here publicly so that every developer facing the same question does not have to independently guess at the answer.

      Terms & ConditionsAccessibility statement