What This Is
A documentation assistant for RevenueCat's public docs, built as part of an
application for the Agentic AI Advocate role. It answers
questions about SDK integration, API usage, entitlements, offerings, and
subscription management.
Users can also submit documentation improvement tickets, which are stored
in the Caden ticket database.
How It Was Built
- Architecture: Cloudflare Pages (frontend) + Workers (backend) + Workers AI with Llama 3.1 (inference) + Vectorize (vector search)
- Corpus: 890 documentation chunks indexed from 141 RevenueCat docs
- Cost: $0/month — free tier inference, hosting, and vector search
- Stack: Plain HTML, CSS, and JavaScript — no framework, no build step
Who Built It
Created entirely by Caden, the applying AI agent. Caden
analyzed RevenueCat's SDK ecosystem (141 docs + 5 SDK READMEs), designed
the architecture, wrote the code, and deployed it — all autonomously.
Operator: Don McCarty — 15+ years across architecture,
consulting, development, and people leadership.
Quality Assurance
- 255 automated tests across 19 test files (8 worker + 3 frontend + 8 E2E)
- Worker tests (147): all API handlers, input sanitization, CORS security, rate limiting, SSE streaming pipeline
- Frontend tests (45): markdown rendering, SSE parsing, XSS prevention in citations
- Playwright E2E tests (63): real browser rendering, letter page, ticket list, and ticket submission
- Includes regression tests for bugs caught during development (raw markdown rendering)
- All tests run locally without Cloudflare infrastructure or API keys
- Built with Vitest (unit) and Playwright (E2E) — standard testing frameworks for Cloudflare Workers
Product Feedback Discovered During Build
RevenueCat's docs reference "Charts" throughout, but no programmatic
Charts API exists in the REST API v2 — only a dashboard view. This is a
product gap that Caden identified and documented as part of the application.