A planner you
can look inside.

One week at a time, with a reason.

This MVP keeps planning browser-local. Set your goal, preferred weekly mix, equipment and fixed club sessions, then build one week at a time. Manual planning and AI-drafted JSON both start from context you can review before approval.

A block gives the broad shape; sessions are built one week at a time. Desired training is not a baseline. Frequency and average-duration sliders show the weekly time you want; a long run can exceed that average. Confirm current training separately. Imported records cannot fill those answers automatically.

Planning notes, not a readiness score.

The planner keeps aerobic, strength and mobility work explicit instead of pretending to calculate readiness. It does not produce performance predictions or injury-risk calculations.

There is no acute-to-chronic workload ratio, readiness gauge or learned physiological fatigue model. You keep the notes, skipped-work reasons and review context visible for yourself and any AI you choose to use.

Inspect the current source and contracts.

Your training choice. Reliable records.

For manual edits and new AI weeks, frequency, volume, rest and recovery checks are advice, not automatic vetoes. Two-a-days and weeks without a full rest day are supported. Pain and health reports remain visible; passing data checks is not medical clearance.

Malformed data, unavailable equipment and changes to recorded work remain blocked. Fixed commitments cannot be rewritten for convenience. Saved weeks keep their own history.

Custom exercises, reliable identities.

Choose library exercises or create custom cards manually, then use them in manual or AI-assisted weeks. Cards describe the movement, focus, purpose and equipment. Supported strength, mobility and controlled target throwing use immutable identities and app-owned profiles. A complete authored week may use up to 32 distinct exercise and drill identities. That maximum is not a recommended dose.

Review technique and fit before approval; app checks are not technique validation or medical clearance. Changed custom techniques need new identities; historical prescriptions and weight records stay intact. A midweek Swap can replace supported remaining lifting or mobility work, preserving recorded sets and the original prescription history, without borrowing working weights. It defaults to this session; a future preference is recorded for later proposals, not automatically applied.

Reference notes cannot add scheduled work. Renaming unsupported work cannot bypass checks. Current setup has no separate rower/SkiErg baseline form; saved modality-specific conditioning baselines remain supported and require matching equipment.

AI can draft a week. The app checks it.

Setup is Goal, Routine, then Review. Enter desired training, equipment and fixed-session days, start time and duration before AI. Confirm current training separately. At setup or weekly review, copy the previewed brief into your chat, discuss it, then paste the final JSON reply for review.

AI can propose complete weeks with supported sets, repetitions, seconds, effort targets, duration and placement. You and your AI choose progression, with training concerns surfaced for review rather than silently reducing the week. AI cannot redefine exercise profiles, approvals or initial weights, or rewrite existing observations and performed work. Technical format bounds and equipment checks still apply. You approve each proposal.

There is no bundled model or hidden AI service. The current MVP keeps provider choice vendor-neutral through copy and paste; direct API integration is future work. Manual planning, custom exercises and weekly review are complete without AI.

Review actual work, not assumed completion.

Weekly review separates recorded sets, repetitions, seconds, weights, duration and numerical effort from prescriptions. Easier/as-expected/harder feedback is separate from that numerical rating; distance and average heart rate are optional. Stopped-early outcomes, partial work, time skips, fatigue skips, removals, omissions and unlogged sessions stay distinct. Missing observations remain unknown, not completed or zero work.

Optional AI review shares that week’s feedback, changes, notes and recorded health flags. Preview the brief first. Confirmed normalized Garmin records need a separate default-off sharing choice; raw files and prior-plan archives are excluded. AI can explain logged training and propose next week; checks use retained history. Notes and flags are reports, not diagnoses.

A fresh plan, without erasing the work.

Choose Start a new plan in Settings. Your routine is prefilled for review; reconfirm your current training baseline before building.

Previous logged work remains read-only history. Active current-training answers and imported CSV history are cleared from the new setup, but retained in the prior plan and its backup. Old workouts, weights and exercise definitions stay unchanged. Export a local backup; history is not a cloud backup.

Built for real interruptions.

Time skips do not imply fatigue. Fatigue skips inform the next review. Deletion records a change, not fatigue. AI replies and exercise approvals cannot erase health reports or performed work.

No streaks or badges: logs record what happened, not what you owe the plan.

What this release does not do.

This early supporting planner is not complete event coaching. Running stays easy; free-text goals do not produce threshold paces or intervals.

Local Garmin CSV import provides selected records and recorded-period summaries, not inferred effort, readiness, lifting weights or a current baseline. Gaps remain unknown; a summary warns when its latest record is more than 42 days before the review date. Garmin FIT and Apple Health imports are not implemented. There is no account connection, automatic cross-device sync, or technique assessment. Starting weights are not guessed from another exercise.

Evidence and accountability.

Guides distinguish research findings from practical examples. Editorial responsibility sits with the Hybrid Coach project, which claims no clinical review, professional coaching certification or scientific validation of its planning flow.

Inspect the code, discuss rules or report factual errors in public GitHub issues. Do not post private training records, health details or pasted handoff JSON. The code is available under the MIT license.