Understand how promoted queries become a portable SQL skills library
A saved query compiles, per role, into a skills library for Claude Code and Codex. The founder walks the model in 79 seconds. No live export on camera.
Chion compiles queries your team saved into skills for Claude Code and Codex. The founder describes that export format in this recording.
Problem. Most text-to-SQL tools lock your queries inside their UI. What Chion does. The founder walking through how saved queries compile into skills for Claude Code and Codex, and what the bundle contains.
Published ·1:19
Want a walkthrough on your own schema?
The founder explains how Chion compiles saved Postgres queries into a per-role SQL skills library for Claude Code and Codex. 79-second video.
What this demo shows
The founder explaining the framework: how Chion reads an existing query corpus (dashboards, ad-hoc reports, ETL scripts) and produces two artifacts, a compiled skills library and an in-product reasoning surface, from the same saved SQL. No live export is performed on camera.
Compiled per runtime, not once for everything
The export is plain Markdown: CHION.md, CONNECT.md, and a skills/ folder holding one SKILL.md per role. Claude Code and Codex each read it from their own path with their own database connection, so the artifact travels while discovery and execution stay runtime-specific.
Outcome predictability over token count
Our SQL reasoning framework is optimized for predictable outcomes, not raw token efficiency. The thirty-thousand-token ceiling per request is dominated by deterministic context gathering, not generative drafting; which is why answers stay consistent across runs.
You own the artifact
The compiled skills library is plain Markdown you can export and put under version control. The query logic stays readable outside Chion, and each runtime you point it at still needs its own database connection.
Read transcript
Recorded May 9, 2026, excerpt. The page copy above carries the current product vocabulary.
00:07Chion converts your existing queries, workflows, and organizational context into a durable business-logic layer that your company owns outright.
00:30Delivery happens through two complementary surfaces.
00:33First, a portable Skills library. Every query is enriched with three things: the business question it answers, the transformation it is permitted to perform, and the role-and-team context that surrounds it. Skills are scoped per team, exportable at every layer, and fungible with any LLM environment that consumes the Skills format.
00:50Second, Chion Studio: our AI-native dashboarding layer that runs the same library through our proprietary SQL reasoning framework. The framework is optimized for outcome predictability, so tokens are not wasted writing SQL from scratch on every request.
01:15Visit chion.ai and contact us to begin testing today.
Frequently asked questions
Common questions about Chion demos and the underlying product.
How is Chion different from other text-to-SQL tools?
Chion enforces three invariants most SQL chatbots skip: every query is read-only SELECT (no writes), every query passes a two-layer validator before execution, and credentials are stored in an AES-256-GCM encrypted vault, never cached in application memory.
Can I use Chion Skills inside Claude Code or Codex?
Yes. The compiled bundle is plain Markdown, and you keep the artifact. Each of those runtimes reads it from its own path and needs its own database connection, so the file travels while discovery and execution stay runtime-specific.
What databases are supported today?
PostgreSQL is the production target. The demos use PostgreSQL 14 fixture databases. MySQL, BigQuery, and Snowflake are on the roadmap.
Does Chion ever write to my database?
No. All generated SQL is SELECT-only with LIMIT enforcement at the validator layer. INSERT, UPDATE, DELETE, and DDL are blocked in code, not by LLM instruction.
Can I run a demo on my own schema?
Yes. Connect your PostgreSQL database with read-only credentials and start asking questions. The trial is 7 days.