Skip to content

Praxis by rustify

Ask your warehouse questions. Get answers you can check.

Praxis builds a semantic layer on your SQL warehouse, answers plain-English questions through queries checked against it, and runs the data work you repeat in numbered steps, with a person checking each one.

Get started How it works

One minute in the portal, against a retail warehouse: a question becomes a checked Cube query and a chart, then the model it was checked against, then a four-step Flow Run.
The AI never writes SQLIt writes a Cube query that is checked against your model before it reaches the warehouse. A wrong name comes back with the nearest real one.
Every step is on screenThe model search, the query, any rejection and retry, the SQL Cube made, and the rows — not just the answer.
Repeated work, written downA Flow is a folder of numbered steps. Rows go to files, not to the AI, and a person reads each step before the next.

How it works

flowchart LR
    U[You, in a browser] --> P[Praxis portal]
    P -- Cube JSON --> S[model store + Cube]
    P -- Python --> X[sandbox<br/>no database password]
    S -- SQL written by Cube --> W[(your warehouse)]
  1. Connect almost any SQL warehouse — Postgres, Snowflake, BigQuery, Databricks and more.
  2. Build a semantic layer on it: what each table means, how tables join, which numbers can be added up.
  3. Ask. Each question becomes a Cube query, checked against the model before it runs. The AI talks to Cube; only Cube talks to the warehouse.
  4. Repeat. Keep questions on dashboards, and turn work you do every week into a Flow.

Find your way

Get startedInstall, connect a warehouse, build a model and ask your first question. Five steps, in order.New here User guideQuestions, dashboards, running a Flow, changing the model, and sharing tools with other agents.Using the portal Flow authoringFolders, phase.yml, prepared queries and scripts, and every check that fails a Phase.Writing Flows OperatingThe two deployments, every setting, security, upgrading, and troubleshooting by error message.Running it ArchitectureLayering, the query gate, the event stream, the Flow engine, storage, and the decisions behind them.Working on it ReferenceEvery HTTP route, the MCP tools, and the glossary of what each word means.Integrating