Hi HN — maker here. I spent 20 years in edtech (co-founded Open English, which hit $171M revenue) and kept hitting the same wall: AI agents that teach need deterministic, structured content, not generated text.

The Orb Platform is three APIs that work together:

Word Orb — 162,251 words, 47 languages, 846K translations, 240K audio files. One GET returns definition, IPA, etymology, translations in native script, and age-appropriate variants.

Lesson Orb — 226,725 structured lesson blocks. Each follows a 5-phase model (hook, story, wonder, action, wisdom) across 10 teaching archetypes, 3 age groups, 19 languages.

Quiz Orb — 21,900 assessment items across 6 question types. Every question is linked to specific lessons and words through a knowledge graph (30,288 edges).

Architecture: Cloudflare Workers + D1 (SQLite at edge) + R2 for audio. No origin server. Sub-5ms edge responses. Same input, same output, every time.

The knowledge graph is the interesting part — it connects words to lessons to quiz items, so an agent can run a complete teach-then-test cycle where the assessment is guaranteed to align with what was taught.

The whole thing is also an MCP server (19 tools): https://mcp.thedailylesson.com/mcp

Playground (no signup needed): https://word-orb-api.nicoletterankin.workers.dev/playground

GitHub: https://github.com/nicoletterankin/word-orb

Happy to answer questions about the edge architecture, the knowledge graph design, or how the archetype system works.