What this is. A personal RAG system for explaining the Cloudflare Workers platform. Every answer is grounded in real, cited excerpts pulled from developers.cloudflare.com and github.com/cloudflare — the model answers only from the retrieved context, not from memory.
Retrieval. D1 (SQLite FTS5, BM25-ranked) over 859 indexed pages across 16 Cloudflare products. Corpus source: developers.cloudflare.com's own llms-full.txt exports per product, plus @cloudflare/workers-types. Chosen over Cloudflare AI Search (the current managed RAG product) as the shippable-today option — see wrangler.toml for the documented upgrade path (not a capability gap: wrangler already has ai-search write/run scopes).
Generation, everywhere on this site. Cloudflare Workers AI, @cf/zai-org/glm-5.2 (Z.ai's agentic/reasoning model), called through the native env.AI.run() binding — no external API key, billed and metered entirely inside this Cloudflare account.
Generation (visualize → diagrams). Same model, used two ways in sequence: first with real function calling (a search_docs tool GLM-5.2 can call to actively explore the indexed corpus, rather than fixed keyword retrieval), then a second call to produce a {svg, explanation} pair from what it found. Every generated SVG is sanitized server-side (no <script>, no event handlers, no <foreignObject>) before it reaches a browser.
Live stats. The charts on the visualize tab are Chart.js reading this Worker's own /api/stats — real corpus data, zero extra cost or dependency.
Source repos. workers-sdk · cloudflare-docs · workerd · workers-types (archived — types now generate from workerd) · workers-rs.
Sister sites. subagentjobs.com · coworkers.subagentknowledge.com.