010: Dynamic models II — smart generation, LLM enrichment, discover endpoint

[acmeguy]

Summary

• Smart model generation: Table profiling with partition detection, field processors, YAML/JS output, and model diff/merge with AI metric preservation
• LLM enrichment: AI-powered metric descriptions, measure suggestions, and schema validation against Cube.js compiler
• Filter builder: Column-level value filtering during profiling, plus a new /column-values endpoint for distinct value retrieval
• Discover endpoint (GET /api/v1/discover): Returns all datasources and cubes available to a WorkOS-authenticated user, with JIT auto-provisioning (user → team → membership from JWT partition claim)
• Hasura actions: GraphQL action definitions for the smart generation flow
• Integration tests: StepCI workflow for smart_gen flow

New endpoints

Method	Path	Auth	Purpose
GET	/api/v1/discover	WorkOS JWT	List datasources + cubes for authenticated user
POST	/api/v1/column-values	Hasura/WorkOS	Retrieve distinct values for a column
POST	/api/v1/smart-generate	Hasura/WorkOS	Smart model generation with profiling

Test plan

• 13 unit tests for discover endpoint (extractCubes, buildDiscoverResponse)
• Verify discover endpoint with live WorkOS token against dev stack
• Run StepCI smart_gen_flow integration test
• Test smart-generate end-to-end with ClickHouse datasource
• Verify LLM enricher with OpenAI API key configured

🤖 Generated with Claude Code