exuvia documentation

exuvia is a workbench for data work that is a graph of plain Python and SQL files. Start with the tutorial, then use the rest as reference.

Getting startedInstall exuvia, build a three-node graph, put a view on it, and turn on the agent. About ten minutes.Start the tutorial

Build

What goes in a project: nodes, parameters, views and the agent.

  • NodesPython and SQL node files, how dependencies are read, and shared helpers.
  • Run parameters and replaysDeclare run parameters, keep one output per value, and replay date ranges.
  • ViewsYour own interface over node outputs, written as front-end code: React and TypeScript files, sandboxed and read-only.
  • The agentThe agent's tools, model providers, project notes and sessions.

Run

How exuvia decides what to run, and where it runs and stores things.

  • Execution modelContent-addressed outputs, derived freshness, schedules, cascade and revisions.
  • Command lineRun and inspect a project without the UI, in CI or from cron.
  • ArchitectureThe stateful control plane, stateless executors, and the executor protocol.
  • DeployingThe executor image, external schedulers, auth, and every environment variable.
  • RecordsWhere run history, schedules, sessions and code revisions are stored.
  • HubServe every project under one directory from a single process.

Integrate

Drive exuvia from another app.

  • HTTP APIEvery route the UI uses, plus Arrow streams and stored Parquet files.

Project

Where the project stands.

  • StatusWhat exuvia does not do yet.