Project Description

Momentum transforms productivity from a static to-do list into a generative construction site — removing the friction between intention and execution. To fully appreciate the “construction site,” you’ll have to see the UI :)

---

“Dump” Mode

Instead of manually organizing tasks, users enter Dump Mode and rapidly unload everything on their mind — messy thoughts welcome.

• No formatting
• No prioritizing
• No structure required

Once submitted, Claude becomes the planner. It:

• Interprets the brain dump
• Decomposes it into structured tasks and subtasks
• Performs relevant research
• Enriches each step with useful context
• Infers metadata such as time estimates, priority, and potential due dates

This transforms cognitive noise into an actionable, buildable plan.

The UI itself is generated from this structured reasoning output.

---

“Do” Mode

Momentum doesn’t just store tasks — it adapts to your moment.

In Do Mode, the interface reshapes itself based on real-time context:

• How much time you have
• What device you’re using
• Where you are
• Any constraints (e.g., can’t talk, low focus, etc.)

Rather than forcing users to manually sort or prioritize, Claude evaluates this context against the structured task graph and recommends the most achievable task right now.

• Waiting for the subway with 10 minutes? → You’ll see something small and phone-friendly.
• Lazy Sunday afternoon at your laptop? → You’ll see something deeper and more strategic.

Once selected, the task view breaks the goal into digestible, never-overwhelming subtasks that Claude has already researched and prepared.

Each subtask expands into a focused workspace with:

• Embedded AI assistance
• Optional agent chat for clarification, iteration, or deeper research

---

Generative Progress

Momentum doesn’t use static progress bars.

As users complete subtasks, a dynamically generated home is constructed on the right side of the screen — visually assembling piece by piece.

When the task is complete, the structure becomes a permanent badge in the user’s profile.

• Progress becomes embodied.
• Momentum becomes visible.

---

How We Shattered UI Conventions

• No static checklist-only interaction
• No chat window as the primary surface
• No manual filtering and selection
• No AI bolted onto CRUD — AI reasoning drives layout, state transitions, and interaction flow
• Fun construction visuals as the user makes progress :)

Claude isn’t a sidebar assistant.
It is the engine shaping what the user sees and can do next.

---

Stability & Architecture

Despite its generative interface, Momentum is architected for stability and resilience.

• Claude outputs structured JSON validated via Zod schemas before any state mutation
• PostgreSQL stores canonical task graphs and metadata
• UI state derives from persisted structured data — never raw LLM text
• Agent sessions are cached in Redis with TTL for reliability and performance
• Type-safe API boundaries prevent drift between reasoning and UI

This ensures:

• Reproducibility
• Predictable state transitions
• A stable working prototype

---

Technologies Used

• Next.js 15 (React 19 + TypeScript) with App Router for a reactive, component-driven UI that re-renders dynamically from structured AI outputs
• Tailwind CSS 4 + shadcn/ui (Radix UI primitives) + Motion for modular, state-driven design with rich animations
• PostgreSQL via Prisma ORM for persistent task graphs, brain dumps, and metadata
• Upstash Redis for agent session caching with TTL
• Supabase Storage for file uploads
• NextAuth.js v5 (Google OAuth + password auth) with Prisma adapter for authentication
• Anthropic Claude API (Claude Sonnet 4.6) for hierarchical decomposition, contextual reasoning, tool use, and web search — with structured Zod parsing
• tRPC + TanStack React Query for end-to-end type-safe API calls, client caching, and server state management
• T3 Stack foundation with validated environment boundaries

N/A