The AI analyst that shows its SQL.
Your database never moves. Connect it, or hand Chion the queries your teams already own, and two things unblock at once. A skills engine for every person, every department, and the whole company. And plain-English answers with the SQL underneath, in minutes, with no migration, from the app or over MCP.
Chion reads with the read-only role you provide and never moves a row.
Connect your database, ask a question, and see the SQL behind the answer.
One recording, start to finish: connect, compile skills, ask, and read the SQL.
Connect. Your database never moves. Point Chion at your tables, or hand it the queries your teams already own and manage.
Compile. Every reviewed query becomes a skill, one per person, rolled up by department and company. Every question you ask and every query you review is added to the engine.
Ask. Business answers and insights in minutes, with the SQL underneath, in the app or over MCP inside Claude and ChatGPT. No migration.
A fast answer nobody can audit is a slow one.
Dashboardsgostale.
AIguessesatyourschema.
Analystsbecomethequeue.
Chion starts every answer from SQL a person on your team reviewed, and shows you the statement it ran.
Every chart carries the SELECT that produced it.
Code-validated as read-only before execution, printed beneath the chart, and re-runnable in any Postgres client.
Generation starts from SQL a person approved.
Each question gets a newly generated SELECT anchored on a query your team reviewed in Studio, so join keys and column names come from checked SQL, not a guess.
A skills engine per person, per department, and for the whole company.
Skills are user-based. Each person's reviewed queries compile into their own skill file, roll up by department, and export as one bundle for the entire company, to run in Chion, Claude Code, Codex, or over MCP.
Nothing migrates. Nothing moves.
Connect the database you already run, or hand Chion the queries your teams already own and manage. Answers within minutes, no migration, no pipeline.
The model sees summaries, never your tables.
Column names and types, statistical profiles, sampled values, and the rows your query returned. Never credentials, never direct table access.
What about my PII? We have a solution for you.
Private GPUs and open models can run off the grid on infrastructure you control, or through secure off-the-shelf providers, arranged under an Enterprise contract, so the model that reads your data runs where you decide.
Thirteen phases stand between your question and your database.
A live skills engine wired to our proprietary semantic generator and RAG database. Schema profiling, a typed SQL contract, two validators, then execution. Same question, same SQL.
Read-only is a policy, not a setting.
Every SELECT is row-capped, scoped to your connection role's RLS policies, and rejected by two code validators if it is anything other than a read-only SELECT. We connect with the role you give us. Your data stays in your database.
Encrypted credential vault
Encrypted at rest. Decrypted in memory only. Never logged.
Read-only by default
SELECT-only permissions. We will never modify or delete your data.
Row-level security honored
Every query runs with your connection role's permissions.
Managed cloud today. Dedicated GPU compute and on-premise model hosting are Enterprise roadmap items discussed under contract.
How it works
Four steps from a question to a live skills engine your team owns.
Connect your data
Point Chion at your PostgreSQL database with host, database, username, and read-only credentials. Or hand it the queries your teams already own and manage.
Semantic Layer
Automatic discovery
Every table, column, and relationship is profiled the moment you connect. No manual schema explanation.
Ask a question about your data…
Ask your data
Ask a question, get a chart. The SELECT that produced it is code-validated before execution and shown beneath the chart.
Generate your real file from Studio. A query you saved and reviewed becomes a skill. Learn more →
Review a query, keep it forever
A query your team reviews and promotes becomes a reusable skill in a personal CHION.md, one per employee, governed by your database permissions. Every promotion adds to the skills engine.
Export the bundle and Claude Code or Codex inherit your team's analytics know-how.
Built by an analyst who got tired of waiting.
Founded in 2025 by Jonathan Dag, former Meta, Twilio, and American Express analyst.Read allJonathan Dag, Founder
After a decade building data infrastructure across Meta, Twilio, American Express, MasterCard, and Bosch, the pattern was always the same: analysts buried in SQL, business users waiting days for a dashboard update, and AI tools hallucinating the answer.
Chion is the analytics layer that should have existed from the start. Ask in plain English, get a code-validated answer backed by real, traceable SQL.
The same standard drives the other product I build, Lensmark, an AI landmark identifier that shows its sources. Photograph a building instead of querying a table, and the answer still has to arrive with the record it came from.
That's the tool I kept wanting someone else to build.
Start in the cloud. Scale on your terms.
Quick Start
Connect PostgreSQL with read-only credentials. Ask a question. Get a chart with the SQL beneath it.
- Managed cloud, nothing to run
- Connect tables or bring the queries your teams already own
- Charts, insights, and the SELECT underneath
- Export skills to Claude Code or Codex
Custom Setup
Configure Chion around your stack. Same pipeline, same validation.
- Your own models, any cloud or open source
- Managed cloud today; dedicated GPU compute and on-premise model hosting are Enterprise roadmap items discussed under contract
- Your own admin: plans, seats, and access controls
Frequently asked questions
Browse the FAQ: security, pricing, pipeline architecture, and integrations, from how SQL is code-validated to how credits and plans work.
Explore Chion
Solutions
Learn
Last reviewed: September 7, 2026







