Your AI-powered mechanical keyboard build advisor.
Describe the sound and feel you want — get a complete, compatible build with real products and real prices.

Try the Live App  ·  Watch the Demo

The video was rushed and doesn't do the app justice — we'd highly recommend trying it yourself.
No sign-in required — Pro access is fully unlocked for everyone, so test as many times as you'd like.

Nia was instrumental to both building and running Switchy. During development, we relied on Nia to index and search the Convex, Clerk, Anthropic, and Next.js docs — it became our primary way to navigate unfamiliar APIs and resolve integration issues quickly across the stack.

In production, Nia is deeply embedded in the product itself:

• Build intelligence — Before Claude generates a build recommendation, Nia's universal search enriches the AI context with community reviews, Reddit sentiment, and real vendor pricing. Recommendations are grounded in what the community actually thinks, not just spec sheets.
• Product validation — When the AI suggests a product outside our local database, Nia verifies it exists, finds a vendor URL, and extracts pricing. This is how we guarantee zero hallucinated products.
• Cross-catalog search — The /search page runs Nia universal search across local products and external community sources in a single query, with 24-hour caching and query deduplication.
• Glossary chatbot — When users ask follow-up questions about keyboard terminology, Nia surfaces relevant community explanations and source material to augment the AI response.
• Automated weekly intelligence — Nia Oracle runs as a cron job that researches new product releases, trending picks from Reddit and YouTube, and group buy status updates — feeding directly into our weekly digest so the platform stays current without manual curation.

Nia gave us something we couldn't have built ourselves in a hackathon: a bridge between a static product database and the living knowledge of an entire hobbyist community.

The mechanical keyboard hobby has a notoriously steep learning curve. New enthusiasts face a maze of 100+ switch types, incompatible parts, inconsistent vendor inventories, and community jargon ("gasket mount", "POM plate", "thocky") that makes entry daunting. Existing resources are scattered across Reddit threads, Discord servers, and spreadsheets — there's no single place where someone can go from curiosity to a complete, compatible build.

Switchy turns this fragmented experience into a guided, delightful one.

Switchy is a full-stack consumer platform where users discover, design, and customize mechanical keyboard builds — with AI doing the heavy lifting of compatibility matching and recommendation.

A full Three.js-powered design studio where users visualize their keyboard in real-time — swap case materials, keycap profiles, colorways, lighting, and camera angles. Designs are shareable via URL.

Browse 85+ switches, 105+ keyboards, 46+ keycap sets, and 82+ accessories with advanced filtering by type, sound profile, brand, and price. Side-by-side comparison mode for switches. Every product links to real vendor pages with real pricing.

A cross-catalog search powered by Nia AI — query across switches, keyboards, keycaps, and external sources simultaneously.

The keyboard community runs on group buys — limited-run products that take months to fulfill. Switchy tracks 46+ active group buys with status timelines, budget management, and a newcomer guide to demystify the process.

A searchable glossary of 80+ terms with difficulty levels, cross-references, and an AI chatbot — so users learn as they browse.

src/
├── app/                    # 22 routes (Next.js App Router)
│   ├── advisor/            # AI multi-turn chat
│   ├── wizard/             # 6-step guided builder
│   ├── builder/            # AI + custom build modes with 3D preview
│   ├── studio/             # 3D keyboard designer
│   ├── search/             # Unified cross-catalog search (Nia AI)
│   ├── switches/           # Explorer + detail + compare
│   ├── keyboards/          # Explorer + detail
│   ├── keycaps/            # Explorer + detail
│   ├── accessories/        # Explorer + detail
│   ├── group-buys/         # Tracker + listings
│   ├── glossary/           # Searchable terms
│   ├── builds/             # Saved builds + public sharing
│   ├── pricing/            # Subscription tiers
│   └── account/            # User profile
├── components/             # 114 components
│   ├── ui/                 # Design system (Card, Badge, Modal, Tabs, Toast...)
│   ├── 3d/                 # Three.js scene, keyboard model, keycap geometry
│   ├── detail/             # Shared product detail (layout, breadcrumb, specs)
│   ├── studio/             # Studio controls, overlays, presets
│   └── builder/            # Build workflow (questions, customizer, pickers)
├── hooks/                  # 10 custom hooks (filters, undo/redo, subscriptions...)
├── lib/                    # Shared types, utilities, constants
└── data/                   # Seed data (85 switches, 105 keyboards, 80+ terms...)

convex/                     # 45 backend files
├── builds.ts               # AI build generation (Claude API)
├── buildAdvisor.ts         # Interactive build advisor
├── stripe.ts               # Payment processing
├── affiliateLinks.ts       # Revenue tracking
├── groupBuyListings.ts     # Group buy management
├── niaSearch.ts            # Nia AI-powered search
├── weeklyDigest.ts         # Automated digest emails
└── ...                     # Auth, CRUD, seeding, webhooks

Why this stack?

• Convex gives us real-time reactivity (live group buy status, instant build saves) without writing a REST API or managing websockets. Queries are just TypeScript functions.
• Claude Sonnet powers the advisor and wizard — it understands nuanced preference language ("something between a thock and a clack, not too heavy") and maps it to real product specs.
• Three.js + R3F for the studio — the 3D preview makes keyboard design tangible before spending money, which is critical for a $200-500 purchase decision.
• Clerk handles auth with zero custom code and integrates natively with Convex via JWT templates.
• Stripe for subscriptions with webhook-driven status tracking.

The AI isn't a chatbot bolted on. It's the core product loop:

1. Preference understanding — Claude interprets natural language into structured specs (actuation force, sound profile, budget constraints)
2. Compatibility matching — Generated builds use real products from our database, ensuring parts actually work together
3. Conversation memory — Multi-turn chat preserves context, so users refine builds iteratively ("make it quieter" / "swap the keycaps for something cheaper")
4. Usage-metered — Free tier gets limited generations; Pro unlocks unlimited AI builds + priority generation

Nia is the connective tissue between our local product database and the real-time community knowledge that makes Switchy's recommendations trustworthy rather than generic. It's integrated across four critical surfaces:

1. Build intelligence — Before Claude generates a build, Nia enriches the AI context with community reviews, Reddit sentiment, and real vendor pricing via universal search. When Switchy says "Gateron Oil Kings are creamy," that's community consensus surfaced through Nia — not a hallucination.
2. Product validation & fallback — When the AI recommends a product outside our local DB, Nia's search API acts as the verification layer: confirming the product exists, finding a vendor URL, and extracting live pricing. This is how we guarantee zero hallucinated products in build recommendations.
3. Cross-catalog search — The /search page runs Nia universal search across all indexed sources simultaneously — local products and external community intel returned in a single query, with 24-hour caching and query deduplication for performance.
4. Automated weekly intelligence — Nia Oracle runs as a scheduled cron job that researches new product releases across major vendors, trending community picks from Reddit and YouTube, and group buy status updates. This feeds directly into our weekly digest — the platform gets smarter every week without manual curation.

Without Nia, Switchy would be limited to its static product database. With Nia, it has access to the living, breathing knowledge of the entire mechanical keyboard community — and that's what makes the difference between a recommendation engine and a genuine advisor.

Switchy has three revenue streams built into the product:

• Saved Builds — Users accumulate builds over time, creating a personal library they return to
• Group Buy Tracking — Group buys span months (IC → production → shipping → delivery). Users check back regularly to track status and manage budgets
• Shareable Links — Studio designs and builds generate share URLs, driving organic traffic from shared links back to the platform
• AI Conversation History — Chat sessions persist, so users pick up where they left off
• Glossary Progression — Terms tagged by difficulty (beginner → intermediate → advanced) create a learning arc that grows with the user
• Budget Management — Monthly group buy budgets create a recurring reason to check in and plan spending

Switchy is designed to feel premium — matching the craft that keyboard enthusiasts put into their builds:

• Custom design system with semantic color tokens, layered shadows, and dark/light mode
• Anti-generic guardrails — no default Tailwind palette, no flat shadows, paired display + sans fonts, spring-easing animations
• Mobile-first responsive across all 22 routes
• Real product images only — zero placeholders, ever
• 3D interaction as a first-class feature, not a gimmick

# Install dependencies
npm install

# Set up Convex
npx convex dev --once --configure=new

# Add environment variables to .env.local
# NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY, CLERK_SECRET_KEY,
# NEXT_PUBLIC_CONVEX_URL

# Add to Convex dashboard: ANTHROPIC_API_KEY, REPLICATE_API_TOKEN

# Start dev server
npm run dev

# Visit /seed to populate the database