Closet

Demo

About

**The Problem:** People waste hours deciding what to wear (365 decisions/year) and buy clothes they rarely use (average person uses 30% of their wardrobe). Meanwhile, AI agents that could help are stuck in "suggestion mode" - they can recommend, but they can't act. The barrier? Trust. How do you let an AI spend your money autonomously? **Our Solution:** Closet is your intelligent wardrobe manager powered by AI agents with verifiable trust. It's three products in one: 1. **Smart Wardrobe Analysis:** Upload photos of your clothes. AI analyzes each item (category, color, style, pattern, season) using computer vision. Builds a searchable, visual database of everything you own. 2. **AI Outfit Generation:** Ask "What should I wear to a business lunch?" Get 3 styled outfit combinations using YOUR actual clothes. Gemini generates photorealistic flat-lay previews showing your items combined together. It's like having a stylist who knows your exact wardrobe. 3. **Autonomous Shopping:** AI identifies gaps in your wardrobe ("You need a winter jacket - you have zero outerwear"). Searches across retailers, finds items that match your style, and can purchase autonomously within rules YOU set. **The Innovation - Autonomous Commerce with Cryptographic Trust:** Traditional AI: "Here's what I suggest you buy." (You do the work) Closet: "I bought it for you. Here's the proof I stayed within your rules." (AI does the work) How we make autonomous purchasing trustworthy: - **User Control:** You cryptographically sign spending rules via x402 protocol (e.g., "$200/month budget, max $50 per item"). Smart contracts enforce these limits - the agent literally cannot exceed them. - **Agent Identity:** Each AI agent has a CDP Server Wallet (on-chain identity). Users subscribe via CDP Embedded Wallets (seamless web3 onboarding, no seed phrases). - **Verifiable Actions:** Every purchase creates a cryptographic proof chain: 1. Intent: User signed the spending authorization (AP2) 2. Process: Agent's decision logic ran in verifiable compute (TEE attestation architecture) 3. Outcome: Purchase recorded on-chain (ERC-8004 claims) **Why Fashion First:** Fashion is the perfect testbed for autonomous agents. Mistakes aren't catastrophic (wrong shirt ≠ medical error), feedback is immediate (you wear it tomorrow), and it's high-frequency (lots of data for reputation systems). Once we prove agents can be trusted with fashion, the model scales to travel agents, meal planning, financial advisors - any domain where you want AI to act, not just suggest. **Current Status (48-hour build):** ✅ Full wardrobe digitization with AI analysis (working) ✅ Multimodal outfit generation (working - generates real visuals) ✅ Product discovery with gap analysis (working) ✅ CDP wallet integration for users (working) ✅ x402 payment flows (working) 📍 Triple-Verified Stack architecture (designed, contracts written) 📍 Agent marketplace + reputation system (roadmap) **The Technical Bet:** We built production-grade infrastructure, not a hackathon demo. TypeScript monorepo with proper testing, type safety end-to-end, background job processing, real database schema. The missing features are just features - the hard problems (AI that understands your wardrobe, web3 payments that feel like web2, autonomous actions you can trust) are solved. **Try It:** Upload a photo of clothing → Get instant AI analysis Ask for outfit help → Get styled combinations with visual previews Connect wallet → See autonomous commerce infrastructure ready to go **The Vision:** Closet proves that autonomous agents can operate in the real economy when trust is cryptographic, not assumed. Fashion today. Everything tomorrow.

How it's made

We built Closet as a production-grade TypeScript monorepo proving autonomous agents can be trusted with real commerce. **Frontend:** Next.js 16 (App Router) + React 19 with TanStack Query for real-time updates. Full type safety from API to UI via ORPC (tRPC-like but better for our use case). **AI Layer:** Multi-provider orchestration (Gemini 2.5 Flash Image for outfit generation and wardrobe analysis). Images are analyzed via computer vision to extract categories, colors, patterns, and styles - then stored with metadata for instant outfit combinations. **Blockchain Infrastructure:** - **Polygon PoS** for all transactions (fast finality, low gas) - **CDP Embedded Wallets** for users (seamless onboarding) - **x402 protocol** for payment intents (users cryptographically sign spending rules) **Backend:** Hono API server with ORPC for type-safe endpoints. BullMQ + Redis for background job processing (image analysis happens async). MinIO (S3-compatible) for image storage. **Database:** PostgreSQL with Drizzle ORM for type-safe queries. Better Auth for wallet-based authentication (SIWE). **Hackiest Part:** We generate photorealistic outfit previews by combining your actual clothing items using Gemini's multimodal image generation. Feed it photos of your real clothes + a text prompt, it outputs a styled flat-lay. Way better than just showing product links.