Your pocket AI HYROX coach and race prep planner
FORGE helps HYROX athletes build long-term training blocks, plan weekly sessions, adapt workouts to real gym equipment, calculate race splits, track PRs, and adjust training load from fatigue feedback.
May 19, 2026 — Training platform expansion
This update expands FORGE from a seven-day coach demo into a broader Web MVP+ training product:
- Long-term AI coach foundation: deterministic 4-8 week plan generation with base/build/peak/taper phases, deload handling, readiness volume changes, and current-week generation.
- Train Hub: new
/trainworkspace with the current block, 30+ HYROX workout templates, filters, scheduling, favorites-ready library data, and a manual workout builder. - Coach workspace: new
/coachentry point for long-term plans, current-week plans, and single-session generation, wired to the guarded Coach API for authenticated users. - Race Prep planner: new
/raceplanner for target finish time, station splits, run splits, ROXZONE estimate, manual station adjustments, and saved race plans. - Metrics dashboard: new
/metricssurface for PR weaknesses, weekly load, readiness trend, completion rate, streak, and pain-note history. - PostgreSQL-ready data model: Prisma now targets PostgreSQL and includes long-term plans, training weeks, scheduled workouts, workout templates, race plans, readiness snapshots, and coach generation records.
- Scheduled workout persistence: Train Hub scheduling and Coach week generation now write first-class scheduled workout records while preserving legacy microcycle compatibility.
- LLM production hardening: generation routes now include cooldown, timeout, retry-on-transient-failure, deterministic fallback, and safe coach-generation metadata without raw prompt logging.
- Demo continuity:
/demoremains public and now seeds long-term plan, library, race plan, metrics, and training data without requiring login, database writes, or LLM calls.
Validation used for this update:
npm run db:generate
npm run test:core
npm run lint
npm run build
npm run test:apiTry the app without creating an account:
http://localhost:3000/demo
Hi, I'm Mervyn, a HYROX enthusiast and AI builder.
I built FORGE because I often train while traveling, where gyms may not have HYROX-specific equipment such as sleds, SkiErgs, or wall balls. I wanted a coach in my pocket that could create a structured HYROX plan, adapt workouts to the equipment available that day, track PRs, and adjust training volume based on fatigue.
FORGE is a fan-made prototype. It is not an official HYROX application and does not use official HYROX logos or trademarks. It is intended as a training assistance tool, not a replacement for qualified coaching.
- AI long-term planning: generate 4-8 week HYROX training blocks with phase, focus, volume target, taper, and deload logic.
- AI microcycle planning: generate seven-day HYROX training plans from athlete profile, race date, target time, equipment, PRs, recent fatigue, and long-term phase.
- Train Hub and workout library: schedule training, browse HYROX templates, filter by duration/focus/equipment, and build manual workouts.
- Race Prep planner: calculate run splits, station splits, ROXZONE, projected finish time, and save editable race plans.
- Metrics dashboard: review PR weaknesses, weekly load, readiness trend, completion rate, streak, and red-flag notes.
- Equipment substitutions: replace sled, SkiErg, wall ball, sandbag, rower, bike, cable, dumbbell, or kettlebell work when today's gym lacks equipment.
- Pacing engine: calculate race, easy, threshold, and interval running paces from target finish time and running PRs.
- Readiness state: summarize recent logs into green/yellow/red guidance to reduce or progress training volume.
- Coach notes: explain why each session was programmed and how readiness, race timing, PRs, or equipment shaped the plan.
- Plan adjustments: show why FORGE reduced volume, switched to fallback planning, applied taper logic, or respected missing equipment.
- Workout logging: record block results, notes, total time, and RPE.
- PR tracking: store benchmark results for HYROX stations and use them in planning.
- Bilingual product experience: switch between English and Chinese UI and generated training content.
- Production-oriented safeguards: authenticated APIs, normalized training data, validated LLM output, and deterministic local fallbacks.
| Home | Train Hub |
|---|---|
 |
 |
| AI Coach | Race Prep |
 |
 |
| Metrics | Workout Log |
 |
 |
git clone https://github.com/mtsui-cmyk/forge-hyrox.git
cd forge-hyrox
npm install
cp .env.example .env.local
npm run devAdd your LLM API key and database settings in .env.local.
Useful commands:
npm run test:core
npm run lint
npm run build
npm run test:apiRun npm run build before npm run test:api, because the API smoke test starts the production server from .next.
CONTEXT.mdcaptures the product direction and domain language.docs/STATUS.mdtracks implementation status and next vertical slices.docs/DEPLOYMENT.mdlists production environment variables and deployment checks.docs/adr/records architectural decisions for LLM coaching, substitutions, and database-first state.src/lib/longTermPlan.tsbuilds and validates 4-8 week HYROX training blocks.src/lib/workoutLibrary.tscontains the HYROX workout template library and filters.src/lib/racePlan.tscalculates race-day run/station splits and editable plans.src/lib/metrics.tssummarizes training load, readiness, completion, streak, and weaknesses.src/lib/coachGuardrails.tsvalidates generated seven-day training plans.src/lib/equipmentSubstitutions.tshandles local equipment substitutions.src/lib/runPrescription.tsderives run paces.src/lib/readiness.tssummarizes fatigue/readiness.tests/coach-core.test.tscovers core coaching behavior..github/workflows/ci.ymlruns the GitHub verification pipeline.prisma/migrations/contains the PostgreSQL database migration.
FORGE is a training assistance tool and not medical advice. Consult a qualified coach or medical professional before starting intensive training, especially if you have pain, injury, or health concerns.
This is a fan-made personal project and is not affiliated with, endorsed by, or sponsored by HYROX.
中文说明
你的随身 AI HYROX 训练教练和比赛准备工具
FORGE 用来帮助 HYROX 训练者生成长期训练周期和每周训练，根据当日健身房器械情况调整训练内容，计算比赛分段配速，记录 PR，并根据疲劳反馈调整训练量。
2026 年 5 月 19 日 — 训练平台扩展
这次更新把 FORGE 从 7 天训练 demo 扩展成更完整的 Web MVP+ 训练产品：
- 长期 AI 教练基础：支持 4-8 周训练周期、base/build/peak/taper 阶段、deload、readiness 降量和当前周生成。
- Train Hub：新增
/train，包含当前训练 block、30+ HYROX 模板、筛选、安排训练和手动 builder。 - Coach 工作台：新增
/coach，登录用户会走受保护的 Coach API；demo 模式仍然保持本地生成且不写数据库。 - Race Prep：新增
/race，支持目标完赛时间、8 段跑步、8 个站点、ROXZONE、手动调整和保存 race plan。 - Metrics：新增
/metrics，展示弱项、周训练负荷、readiness 趋势、完成率、连续训练和疼痛备注。 - PostgreSQL 数据模型：Prisma 已切换 PostgreSQL，并新增长期计划、训练周、日程训练、模板库、race plan、readiness snapshot 和 coach generation 表。
- 日程训练持久化：Train Hub 安排训练和 Coach 当前周生成会写入
ScheduledWorkout，同时保留旧 microcycle 兼容。 - LLM 生产加固：生成路由加入 cooldown、timeout、一次瞬时失败重试、fallback 和安全元数据记录。
- Demo 延续：
/demo仍然无需注册、无需数据库写入、无需 LLM，同时包含长期计划、模板库、race plan 和 metrics 示例。
本次验证：
npm run test:core
npm run lint
npm run build
npm run test:api- AI 长期规划：生成 4-8 周 HYROX 周期，包含阶段、重点、训练量、taper 和 deload。
- AI 微周期规划：根据长期阶段、运动员档案、比赛日期、目标时间、器械、PR 和疲劳反馈生成 7 天训练计划。
- Train Hub 与训练库：安排训练、浏览 HYROX 模板、按时间/重点/器械筛选，并手动创建训练。
- Race Prep：计算跑步分段、站点分段、ROXZONE、预计完赛时间，并保存可编辑 race plan。
- Metrics：查看 PR 弱项、周训练负荷、readiness 趋势、完成率、连续训练和疼痛备注。
- 器械替代：没有雪橇、SkiErg、墙球、沙袋、划船机等器械时，生成可执行替代方案。
- 配速引擎：根据目标完赛时间和跑步 PR 计算比赛配速、轻松跑、阈值跑、间歇跑配速。
- 训练状态反馈：根据近期日志输出 green/yellow/red 训练状态，用于调整训练容量。
- 教练解释：解释每天为什么这么练，以及 readiness、比赛时间、PR 或器械如何影响计划。
- 计划调整原因：解释为什么 FORGE 降量、切换 fallback、应用 taper 或尊重缺失器械。
- 训练打卡：记录每个训练块的结果、备注、总时间和 RPE。
- PR 追踪：记录 HYROX 站点个人最好成绩，并用于后续计划生成。
- 中英文体验：界面和生成训练内容支持英文/中文切换。
git clone https://github.com/mtsui-cmyk/forge-hyrox.git
cd forge-hyrox
npm install
cp .env.example .env.local
npm run dev在 .env.local 中填入 LLM API key 和数据库配置。
无需创建账号即可体验：
http://localhost:3000/demo
FORGE 是训练辅助工具，不是医疗建议。进行高强度训练前，尤其是在有疼痛、伤病或健康风险时，请咨询专业教练或医生。
本项目为个人爱好作品，非 HYROX 官方应用，也不代表 HYROX 官方。