PACER · AI Running Coach
A coach for every kilometer you run. Your condition changes — so does your plan.
PACER is an AI-native running coach — an AI agent that makes its own decisions, not a chatbot wrapper. Via the official COROS MCP it securely reads up to about two years of your training history, and builds on Jack Daniels' VDOT, Hansons cumulative fatigue and 80/20 polarized training to craft — and continuously adapt — a training plan for your next race. Your account password is never stored.
A three-layer scientific decision model
No guesswork — a purely data-driven personal coach. On every sync the model re-reasons, so your plan is always the best answer for right now.
- Multi-dimensional state sensing: reads body and environment signals — COROS data sync, temperature and environment adjustments, sleep and HRV recovery monitoring.
- Core computation engines: decisions made with sports-science models — Daniels VDOT pace assessment, Banister TRIMP fatigue quantification, runner injury-medicine risk alerts.
- Dynamic plan output: executable daily training — weekly adaptive plans, pre-race taper, precise hydration and nutrition guidance.
Why PACER is an AI agent, not a chatbot wrapper
A plain LLM can chat, but it also miscalculates and makes things up. PACER wraps it in a self-built core that makes it a disciplined coach:
- Scheduling orchestrator (safety rails): say "I want to run more this week" and it won't just pile on volume — anything beyond what your body can absorb is clamped back to science-based limits, and plan changes only take effect after you confirm.
- Kinematics engine (deterministic math): pace, training load and hydration numbers are computed precisely in code with sports formulas — never estimated by the AI.
- Knowledge base (RAG, answers with sources): whether you can run with Achilles pain, how to carb-load before a race — answers are grounded in sports-medicine and nutrition literature, not generated out of thin air.
- Long/short-term memory engine: remembers your old injuries, preferences and goals, distills patterns you haven't noticed (late-run fade on long runs, mid-week dips) and speaks up at the right moment — it gets to know you better the more you use it.
Core capabilities
- One-click COROS sign-in and training data sync (up to ~24 months of history)
- AI-generated personalized race training plans for 5K / 10K / half marathon / full marathon
- Daniels VDOT fitness assessment with E / M / T / I / R five-zone science-based pacing
- Training reviews, in-depth pre-race analysis and finish-time prediction
- Conversational AI agent copilot (orchestrator safety rails + deterministic pace engine + knowledge-base RAG + long/short-term memory), for questions and adjustments anytime
- The three-layer decision logic is open-sourced as a Skill — plug it into Claude, Cursor or other MCP/Skill-capable tools yourself
FAQ
What is PACER?
PACER is an AI-native running coach (web app). At its core it is an AI agent that makes its own decisions — not a chatbot wrapper. It securely connects to your training data via the COROS MCP and builds on Daniels VDOT, Hansons cumulative fatigue and 80/20 polarized training to generate and dynamically adapt your personalized race training plan.
How is PACER different from a regular running app, or a ChatGPT wrapper?
Most running apps record and display data; chatbot wrappers can talk but also miscalculate and make things up. PACER is a disciplined AI agent: around the LLM it runs a self-built scheduling orchestrator (safety rails), a kinematics engine (deterministic pace/load math), a sports-science knowledge base (RAG) and a long/short-term memory engine. After every sync it re-reasons and outputs the best plan, review and pre-race analysis for right now — not a fixed template.
Which devices and data sources are supported?
Currently one-click sign-in via the official COROS MCP, syncing training data from COROS devices (up to about 24 months of history). Your account password is never stored. More data sources are in the works.
What training science is PACER built on?
Daniels VDOT (fitness assessment and E/M/T/I/R pace zones), Hansons cumulative fatigue, 80/20 polarized periodization, with reference to Banister TRIMP fatigue quantification and ACSM nutrition & hydration guidelines.
Which runners and races is it for?
Runners with a concrete race goal: personalized training plans for 5K, 10K, half marathon and full marathon.
Does PACER cost anything?
It is currently free to use.
Can I run this model in my own AI tools?
Yes. PACER's three-layer decision logic is open-sourced as a Skill. In tools that support MCP/Skills (such as Claude or Cursor), point them at the repository (github.com/xiaolouJB/ai-running-coach), connect your COROS data, and you can generate the same training plans in your own environment.
How does PACER protect my data?
Data is read via official COROS MCP authorization; your account password is never stored. AI reviews and advice are based only on your real synced data — when data is missing it says so honestly and never fabricates mileage, pace, heart rate or medical diagnoses.
Further reading: Training methodology (Daniels / Hansons / 80-20) · FAQ
Connect your COROS account and start with PACER →
© 2026 PACER · AI Running Coach · pacer.xiaolou.space