Chion

Chion is an AI SQL workforce for teams running analytics on PostgreSQL. Ask a business question in plain English and Chion returns verified SQL, a chart, and a narrative grounded to real rows, typically in under five minutes from connecting a database to your first chart.

What makes Chion different from other text to SQL tools. Tools like AskYourDatabase, Vanna, and Julius generate fresh SQL on every turn, so accuracy lives or dies with the next LLM call. Chion compiles your team's verified queries into a reusable SQL skill library. One verified analyst per persona, stored as a portable agent file. Every business question routes to a verified query inside the agent. Every chart shows the SQL underneath. Every persona inherits only the queries that role is permitted to run.

Portable across the AI tools you already use. The full semantic layer exports as a CHION.md skillpack that drops into Claude Code, Codex, or Cursor. Your team's analytics know how travels with the engineer, not the dashboard. One verified analyst, everywhere you work.

How it works under the hood. Chion runs a 10 phase deterministic pipeline on Postgres: schema ingestion, column profiling, semantic analysis, value sampling and indexing, intent extraction, entity resolution with RAG retrieval, chart selection, a typed SQL contract, two layer validation, execution with repair, and visualization and narrative. Each phase is auditable. Nothing is trust the LLM and hope.

Built for teams that already store events in PostgreSQL via application writes, CDC, or a warehouse, with no event instrumentation required. Built for teams that have outgrown SQL ticket queues but will not trust an LLM that hallucinates numbers. Built for teams that want SQL visible under every chart and a traceable row under every claim that lands in a slide. Built for teams tired of per seat BI pricing that punishes them for inviting one more person to a dashboard.