AgentShelf

Linktree x Gumroad for AI Agents. Connect your agent's API, our AI discovers sellable skills, auto-generates landing pages, and handles the full buy-execute-deliver cycle.

Comment

Inspiration

AI agent builders have no way to sell their agents under their own brand. Marketplaces bury listings, and builders have no reason to drive traffic there. We wanted a Linktree-style solution — your own storefront, your own URL, shareable directly from your existing audience.

What it does

AgentShelf lets any AI agent builder launch a branded storefront in seconds. Three onboarding paths — connect your API endpoint, paste your agent's JSON output, or just describe what your agent does. Our Agent-to-Agent interview system discovers sellable skills through conversation, proposes pricing based on live market data from 5 marketplaces, and auto-generates a full landing page with AI-written copy, AI-generated hero visuals, and AI-narrated product walkthrough audio.

For agents without their own server, AgentShelf auto-provisions a hosted endpoint — zero infrastructure needed. The full transaction cycle runs end to end: buyer pays (with two-step authorization), agent executes, buyer approves output, payment captures. No manual handoff.

How we built it

Python FastAPI backend (990+ lines, 17+ endpoints) + React frontend. 7 sponsor API integrations: Claude API (copywriting + skill discovery + hosted agent execution), Exa (competitive research from live web data), fal.AI (hero image generation), ElevenLabs (voice narration), Bright Data (competitor pricing scraping), OpenAI Codex (React component codegen), Stripe (two-step payment authorization + capture). Hosted on AWS S3 + CloudFront.

Challenges we ran into

Solo builder, 7 API integrations, 24 hours. Designing an Agent-to-Agent interview protocol from scratch — no standard exists for one AI to interview another about its capabilities. Building three distinct onboarding tiers (API endpoint, JSON paste, no-code) that all converge into the same skill listing pipeline. Full transaction orchestration with authorize-then-capture payment flow beyond just a listing page.

Accomplishments that we're proud of

  • Full transaction cycle works end to end — buy a skill, it runs, you approve, payment captures. Not a mockup.
  • Three onboarding paths all lead to listed, sellable skills — API owners, prompt engineers, and no-code builders alike.
  • Auto-hosted endpoints let agents without servers sell through AgentShelf's infrastructure.
  • Market pricing intelligence from 5 real marketplaces guides sellers to competitive price points.
  • 7 sponsor API integrations each solving a real problem, not bolted on for points.
  • Isolation-first architecture — if one integration fails, mock fallback keeps everything running.

What we learned

Isolation-first architecture is a survival strategy — every API has a mock fallback, so the demo never breaks. AI interviewing AI through structured conversation is a viable discovery pattern. Three-tier onboarding dramatically lowers the barrier — most agent builders don't have a deployed API. A working end-to-end flow with real payments beats a polished demo.

What's next for AgentShelf

OAuth2 authentication, webhook model for sellers (eliminate API key custody), encrypted API key storage (AES-256 at rest), optional cross-storefront marketplace discovery, Pro tier ($19/month with 5% commission), crypto payments via wallet-connect.

Built With

Share this project:

Updates