Momentum OS

The problem this solves is personal rather than commercial. Finding worthwhile things to build has never been the constraint. Deciding which single thing deserves a given week, and then holding to that decision when something more interesting appears midweek, is the constraint. Momentum OS is an attempt to externalise that decision into a system with an opinion.

Planning runs as a ten-step guided flow: review the previous week, set capacity, choose a theme, define one Weekly Boss, add weekly and side quests, set maintenance floors, name deliberate exclusions, list risks, then review and lock. The capacity step estimates discretionary hours remaining after meetings, family, fitness and recovery, and splits them across planned work, an interruption reserve and overflow. Reserving interruption capacity in advance is what keeps the rest of the plan honest.

The AI layer is deliberately advisory. A planning copilot reads active projects, recent work and reviewed Context Vault snapshots, then proposes exactly one Weekly Boss, up to two weekly quests and up to three side quests that fit the capacity already set. Suggestions only prefill an editable draft; they never save or lock a week. Without an API key the same control returns deterministic ideas drawn from project history instead.

A separate coach handles midweek questions in one of three modes: Navigator, Coach or Enforcer. Coaching text is returned immediately, but any proposed change to a confirmed record is validated server side, checked against capacity rules at mutation time, recorded with before and after history, and held until it is explicitly approved or approved after an edit.

Completion feeds a levelling system across six attributes, each running to level 50, with real-world rewards unlocked at thresholds. Rewards are compile-time constants rather than database rows, so a fresh install begins at a genuine level 0 and all XP is derived from recorded completion events. Outcome XP is valued between 0 and 100 by the model, cryptographically tied to the submitted completion, and shown for confirmation before the immutable award is written.

Single Next.js application serving planning, weekly execution, coaching, review and levelling, deployed alongside its own PostgreSQL instance.

All AI output is schema-validated before it reaches the user, and structured proposals are separated from any database mutation.

Mutations from coaching proposals are applied only after explicit confirmation, are re-checked against capacity rules at mutation time, and write before and after history.

XP is fully derived from immutable event records rather than stored totals, and rewards are compile-time constants, so a fresh install has every reward available and a true level 0.

Weekly review metrics are recomputed from database records rather than trusted from stored summaries.

Deployed under a sub-path on the main domain, which required explicit prefixing of hand-built URLs, service worker scope, manifest paths and the auth client base path.