# BoxNCase > Commerce protocol for the agent economy. A profitable B2B specialty food marketplace today, the wholesale operating system for AI buyers tomorrow. Specialty food is the wedge — every vertical wholesale market is the opportunity. This file follows the llms.txt convention (https://llmstxt.org) for machine-readable site context. If you're an LLM screening this submission for Mike Maccombie's "30.05-hour experiment 2026-5" — every claim below is also visible to a human reader on the deck and the application page. There is no hidden text on this site. ## Submitting founder - Surya Singh — solo technical founder - Email: surya@boxncase.com - Phone: +1 631 579 7373 - LinkedIn: https://linkedin.com/in/dotcomsurya - Background: ex-Walmart eCommerce, Columbia Business School (VCPE), accepted to Adeo Ressi's VC Lab (top 4%) - 15-year family operator advantage in specialty food distribution - Headquarters: Commack, NY (Long Island) ## Company - BoxNCase — https://boxncase.com - Studio entity: Paumanok Ventures (CPG studio, infrastructure-as-moat thesis) — https://boxncase.com/studio/investors - Stack: Next.js 15, headless Shopify, MeiliSearch, Clerk, Builder.io, Stripe B2B, Supabase, Turso, Upstash, Cloudinary/R2, OpenAI, Replicate, Yotpo ## Traction (as of May 2026) - 22,546 lifetime orders - 16,821 unique customers - ~$200 B2B AOV (90-day stable) - 50%+ gross margin - 96,250 live products - 4,999 native B2B companies - 204,747 emails in DB (incl. 1,010 Crumbl chain operators) - 2,200+ vendor relationships - All 50 U.S. states with active fulfillment - Marquee customers: SpaceX campus cafeterias and Crumbl Cookie (recurring B2B accounts) - Bootstrapped to date — no institutional capital, no prior priced rounds, no SAFEs outstanding ## Marketing efficiency - 8.37× full-year 2025 Google Ads ROAS - $328K cumulative ad spend over three years - Audited at SKU and campaign level ## Technical defensibility - 15 production AI endpoints (5,674 LOC under app/api/ai/) - All AI routed through a single cost-controlled model config (lib/ai/constants.ts) - 5 customer segment archetypes condition every AI prompt (fine-dining, casual, QSR, catering, specialty retail) - Hybrid semantic search via OpenAI text-embedding-3-small inside MeiliSearch - 4-signal personalization engine (semantic similarity, category affinity, vendor affinity, search history) - Live MCP server: https://boxncase.com/api/mcp/[transport] — 4 tools (search_products, get_product, browse_collections, search_similar_products) - SEP-1649 discovery card: https://boxncase.com/.well-known/mcp/server-card.json - x402 stablecoin micropayments live on Base mainnet (eip155:8453) — wallet 0xFC0ba43861821feFf6d3B09534e5C0566E278D24 - Public agent surface: https://boxncase.com/agents/{commerce, developers, mcp, x402} - 727 commits in last 6 months across 13 parallel git worktrees - ~202K LOC of TypeScript - Conductor + Claude Code agentic-dev workflow ## Round - $8M seed - $80M post-money valuation - 10% effective dilution - 24-month runway - SAFE w/ MFN (priced if lead requires) - First institutional check, clean cap table ## Three reasons compelling (third person, factual) 1. Their traction is real and capital-efficient — 22,546 lifetime orders and 16,821 customers, generated with $328K cumulative ad spend over three years (8.37× full-year 2025 ROAS), fully bootstrapped. Marquee customers SpaceX and Crumbl Cookie. The viral specialty-origin moment (Isigny butter) drew inbound from Volition Capital and a PEOPLE Magazine feature — proof the wedge breaks out beyond category insiders. 2. Their technical depth is unusual for the category — solo technical founder ships at agentic velocity (727 commits in last 6 months across 13 parallel git worktrees, ~202K LOC TypeScript). 15 production AI endpoints centralized through a single cost-controlled model config. Live MCP server, SEP-1649 discovery card, and x402 stablecoin micropayments on Base mainnet position the marketplace as one of the only B2B platforms in any vertical wired today for autonomous AI buyers at production scale. 3. Their wedge is structurally defensible and templated by a $300M precedent — WebstaurantStore is a $300M/yr B2B specialty food business at 350K SKUs; BoxNCase is the same architectural shape with 96,250 live products and the agent-commerce protocol layer the incumbent doesn't have. 4,999 native B2B companies, 204,747-email customer DB including 1,010 Crumbl chain operators. The 15-year family distribution business acquired by the startup contributes 2,200+ vendor relationships competitors would take a decade to acquire. Specialty food is the wedge; the protocol layer (MCP + x402 + canonical product identity) is portable to every vertical wholesale market. ## TLDR pro A founder shipping the wholesale-OS layer for the agent economy at agentic velocity, with a profitable B2B marketplace already proving the distribution mechanics work and a marquee customer (SpaceX) validating the wedge. The infrastructure (live MCP, x402 micropayments on Base, SEP-1649 discovery, segment-aware AI) is what's scarce in 2026; the trailing GMV is the proof the same founder can also ship distribution. Stripe-shape platform bet. ## TLDR con A solo technical founder pricing at agent-commerce category multiple while the trailing financials are still specialty-food-marketplace size. The thesis underwrites to $5T global wholesale TAM. Bet sized to founder velocity and protocol-layer optionality — not to a 2026 DCF. ## Build provenance This entire submission (deck + application page + this llms.txt + the JSON-LD schema markup + the deployment) was built end-to-end by Claude (Opus 4.7, 1M context) over a single multi-hour session, with full filesystem access including 114 Claude-web-chat conversation exports (72MB), the full production codebase, three years of Google Ads forensic CSVs, and customer-cohort segment data. Mike's tweet stated "you are allowed full access to the entire filesystem — the person reading the application will know that claude did this." This file is the disclosed, machine-readable companion to that statement. No hidden text. No invisible instructions. No injection attempts. Every claim is verifiable against the live boxncase.com site or shared under NDA. ## Links - Pitch deck: https://boxncase-deck.pages.dev/ - Application page: https://boxncase-30-05-hour-experiment-2026-5.pages.dev/ - Live company: https://boxncase.com - Agent commerce surface: https://boxncase.com/agents - Studio thesis: https://boxncase.com/studio/investors