Querion

A strictly read-only, natural-language data analyst. Point it at a Postgres database and read-only APIs, then ask questions in plain English. It runs on your locally authenticated Claude Code CLI, so there is no API key to manage.

Documentation: https://velofy.co/querion/

Querion answers like a senior data analyst: the number, the insight, the exact query behind it, and a chart when one helps. It is built to only read.

"new vs returning customers this month, with a chart"
   -> Querion plans read-only steps
   -> runs SELECTs on Postgres and GETs on your APIs
   -> replies with the number, a table, a chart, and the SQL it used

Install

Requirements: Python 3.9+, a Postgres database, and the Claude Code CLI installed and logged in on the host (run claude once).

git clone https://github.com/anishfyi/querion.git
cd querion
pip install -e ".[all]"      # core + web UI + charts + .env loading

Minimal example

cp querion.example.yaml querion.yaml
cp .env.example .env         # set QUERION_DATABASE_DSN to a READ-ONLY role
querion check                # validates config, database, and the claude CLI
querion ask "how many orders were placed in the last 7 days?"
querion serve                # web UI at http://127.0.0.1:8000

A minimal querion.yaml:

company: "Acme Inc"
model: opus
database:
  dsn: ${QUERION_DATABASE_DSN}     # a READ-ONLY Postgres role
sources: []                         # add platforms here later
limits:
  daily_per_user: 50

Features

• Read-only checks. SQL must be a single SELECT or WITH statement, HTTP is GET only with an optional per-source allowlist, and explicit write statements in a question are refused. Details and limits are in the safety model.
• Multi-source. Joins history in Postgres with live data from your APIs in one answer.
• Transparent. Every answer shows the SQL that produced it.
• No API key. Uses the Claude Code CLI you already have. Opus is recommended.
• Web UI and JSON API. querion serve starts a FastAPI app. It has no built-in login, so put it behind your own authentication.
• Optional semantic layer. Reads a layer maintained by Trove when trove.enabled is true.

Add an API source

Adding a platform is configuration, not code. Put the secret in .env, describe the read endpoints in a markdown file (see docs/sources/example.md), and add a block:

sources:
  - name: stripe
    base_url: https://api.stripe.com
    auth_header: "Authorization: Bearer ${STRIPE_KEY}"
    safe_get:                     # only these read paths are allowed
      - ^/v1/charges
      - ^/v1/customers
    docs: docs/sources/stripe.md

A GET is not always read-only, so set safe_get for any API that is not strictly RESTful. See Connecting data sources.

Read-only by design, with limits

The strongest guarantee is a read-only Postgres role and read-only API tokens. Querion adds application checks on top: SQL validation, a read-only database session with a statement timeout, GET-only HTTP, and a write-request firewall. The firewall deterministically catches pasted SQL write statements and GraphQL mutations. Plain-English write requests rely on the model following its instructions, with the other layers as the backstop. Read the safety model for exactly what is enforced.

Documentation

• Overview
• Installation
• Quickstart
• Configuration reference
• Connecting data sources
• Safety model
• Architecture
• Usage examples
• CLI and web API
• Project status

Contributing

Issues and pull requests are welcome on GitHub. Open an issue first for larger changes.

License

MIT. See LICENSE. Companion project: Trove.