// the find
Canner/vulcan-sql
Data API Framework for AI Agents and Data Apps
VulcanSQL lets data teams write templated SQL files and automatically serves them as RESTful APIs — think dbt but the output is HTTP endpoints instead of tables. It targets analysts and backend developers who need to expose warehouse data to apps or AI agents without writing a full API layer. Last commit was July 2024 and the repo has gone quiet since.
The dbt-inspired SQL templating with Jinja-style variables is genuinely useful for data teams who already think in SQL. DuckDB as a local caching layer is a smart call — it reduces hammering BigQuery or Snowflake on every API request. Connector breadth is real: BigQuery, Snowflake, DuckDB, PostgreSQL, ClickHouse, ksqlDB all supported via a plugin system. Auto-generated OpenAPI docs from the YAML schema files means AI agents and frontend devs get a spec without extra work.
The project is effectively abandoned — last push July 2024, 42 forks and 791 stars is thin for a framework asking you to build production APIs on it. The catalog server is a full Next.js app bolted on as a separate package with its own Docker container, which is significant operational weight for what amounts to a data browser. No obvious story for auth beyond what's in the catalog server, so access control on the actual API endpoints falls on you. The Nx monorepo with 20+ packages makes it hard to understand what you actually need to install versus what's optional, and the docs site (vulcansql.com) may not be maintained.