--- title: "NuroStride: your running coach needs context" description: "NuroStride turns your goals, training background and weekly availability into a personal AI running-coach skill package. Why the context belongs with the athlete, and what the software can actually promise." canonical: https://arnekellmann.de/blog/nurostride-your-coach-needs-context published: 2026-10-07 modified: 2026-10-10 lang: en --- # NuroStride: your running coach needs context A runner asks an AI for a marathon plan. The answer has intervals, easy runs, a long run and a taper. It looks complete. But the runner can train only on four particular days, has just returned from a break, and found last week's easy run unusually hard. Unless those facts survive the conversation, next week's answer can quietly forget all three. [NuroStride](https://nurostride.com/?source=portfolio_en) is built around keeping that context explicit. Its onboarding software creates a personal running-coach skill package for the athlete's own AI environment. The athlete gets the coaching instructions, their profile, reusable review workflows and the calculation tools together. The purchase buys a versioned package and the software for configuring it. The athlete uses a separate AI service to work with it. I developed NuroStride from my background as an ambitious amateur athlete and considerable time spent learning about training science. I wanted to bring training data, everyday constraints and scientific methods together in a coaching process that can be examined. My personal experience explains the product perspective; it is not evidence of effectiveness. That distinction matters. A neat plan is an output. A coaching process also needs to remember why the plan was chosen, what actually happened and which assumptions have changed. ## The useful part starts before the plan Consider a hypothetical amateur preparing for a marathon. They want to run under 2:55, have four reliable training days, and can do a long run on Sunday. Their last few weeks have been uneven. Tuesday has room for a longer session; Thursday has forty minutes. There is no recent race result that establishes the proposed marathon pace. A generic prompt can turn the target time into a pace and build the week around it. The arithmetic is easy; the reasoning is the problem. A target pace says what a runner wants to achieve. It does not establish what they can currently sustain. In NuroStride, those are different entries in a shared context: the wish, the recent training history, the available time and the missing evidence. The skill instructs the AI to keep missing data unknown and explain any assumptions. A proposed week should fit the recorded calendar and the training the athlete has actually tolerated. It should not borrow a training history from an imagined runner who already fits the target. This is where individualisation becomes concrete. Not in a greeting with your name, but in a Thursday session that fits forty minutes, a Sunday long run that follows the recent build-up, and a reason for postponing a faster workout when the evidence is incomplete. ## The software produces files you can inspect The onboarding collects the inputs needed to assemble that starting context. Personalisation happens in the browser. NuroStride does not need to receive a runner's training background in order to put it into their package. The server handles the purchase, access to downloads and the versioned base material. The exported ZIP is an artifact the athlete can keep and inspect. It contains the coach skill and supporting files rather than requiring the athlete to leave their history inside a new hosted chat product. Instructions describe how to use the package and maintain the context as training continues.
NuroStride mobile goal selection with three running priorities and optional Sub-2:55 context.
Goal selection in the redesigned mobile NuroStride onboarding, captured on 8 October 2026. Blue selection outlines and separate goal cards guide the inputs; this image contains no athlete data and shows no training outcome.
Browser-local assembly has a precise boundary. When the athlete later gives the files or training records to their chosen AI service, that is a separate data transfer governed by that service. The package cannot make that provider private by itself. The athlete decides what to share and should check their AI account's data settings. Nor does an exported file keep itself current: someone must add the completed sessions and correct outdated context. ## A skill defines a method; a connector provides access An AI skill is a folder of instructions and resources that a supporting AI tool loads for a particular task. The format does not contain a running language model. [Anthropic's skills documentation](https://support.claude.com/en/articles/12512176-what-are-skills) separates that procedural knowledge from MCP, the protocol used to connect an AI tool to external services and data. For a running coach, that means a skill can explain how to review a training week even when no sports platform is connected. A connector could provide access to activities, if the athlete installs and authorises one. It does not by itself decide whether a session was appropriate. Those are two different jobs. The launch package works from context and records the athlete supplies. It does not arrive with an already connected watch or training account. This keeps the starting point understandable: review the instructions, provide the inputs, check the result. Check platform compatibility and any later connector setup against the current installation guidance; the same files do not guarantee identical behaviour in every AI tool. ## Check the calculations separately from the advice The package includes a Python calculator for defined training-load methods and target-time splits. The package documents its calculation rules. Session RPE uses duration and perceived exertion; heart-rate TRIMP needs its explicitly configured parameters. The software should not silently mix those methods or manufacture a missing resting heart rate. A split calculator can tell our hypothetical runner how fast each kilometre would need to be for 2:55. That is arithmetic, not a prediction. Likewise, smoothed load measures describe the supplied history. They do not certify that the athlete is recovered or safe to train. The [IOC consensus on load and injury](https://ostrc.no/globalassets/publications/soligard_2016_bjsm_how-much-is-too-much.-ioc-consensus-statement-on-load-in-sport-and-risk-of-injury.pdf) discusses sport and non-sport stress, individual responses and monitoring. It also reports that a single response marker does not consistently predict injury. That supports treating load as one input among several, rather than making a numerical score the final decision-maker. It does not validate NuroStride or establish its effectiveness for amateur runners. This division gives the athlete something useful to audit. The calculation should reproduce the same value from the same inputs. The AI's recommendation then needs to explain how that value, the actual sessions and the athlete's feedback informed the proposal. ## The next week is the real test of the process Return to our four-day marathon runner. Suppose Thursday's easy run feels much harder than expected and Sunday's long run is shortened. The review should record those facts and ask for the missing context, rather than treating the planned kilometres as completed kilometres. The package's review workflows provide a place for that reconciliation. Training planning, recovery, fueling and race-day preparation share the same athlete context. Changes should state what changed, what evidence supports the change and what still needs to be checked. A later conversation should be able to understand the decision without relying on a reassuring sentence in an old chat. This does not remove the athlete's judgement. AI can misunderstand a session, apply instructions badly or invent an explanation. Pain, illness and concerning symptoms require an appropriate human response; a coach skill cannot diagnose them or grant medical clearance. A professional who knows the athlete can add oversight that a set of files cannot supply. ## What the launch offer includes NuroStride launches at €69 as a one-time purchase of the Running Coach package, with twelve months of package updates included. The first specialist context supports a Sub-2:55 marathon goal within the same core offer. It is a target to work toward and discuss, not an entry promise that every buyer can achieve it. A separate Premium Setup service is offered at €249 by enquiry and appointment. It includes the kit and a bounded technical setup service. It is for someone who wants help getting the package working in their chosen environment, rather than purchasing ongoing training supervision. Scope and appointment are agreed before payment. An AI account, its usage charges and the athlete's devices are separate. The package also has versions: a change to the instructions or calculator can be identified and reviewed. Buying files should not obscure which method you have installed. For the runner at the beginning of this article, the first useful result is modest and tangible. Their availability, history and uncertainties now travel with the coaching instructions. The next plan can be challenged against that context. That is the product NuroStride sets out to deliver: a coaching method the athlete can own, configure and keep accountable as the weeks change.