Skip to main content

Documentation index: llms.txt. This page is also available as markdown: append .md to this URL or send Accept: text/markdown.

Local AI Development

Preview

This feature is currently available in a private preview and may not be accessible to all customers.

Coalesce supports building data pipelines locally with the AI agent of your choice: Claude Code, Cursor, Codex, Snowflake Cortex, or any other AI coding agent. Your pipeline lives as files in a Git repository on your machine. Your agent edits those files, the Coalesce CLI validates and runs them against your warehouse, and Coalesce Desktop gives you a visual view of the pipeline as it takes shape.

You bring your own agent and your own LLM. Coalesce does not proxy or host the model.

The Components

ComponentWhat it doesWhen you need it
coa CLIAuthors, validates, plans, deploys, and runs pipelines from your local workspaceAlways. It is the engine of local development
Coalesce DesktopA visual review surface over your local workspace. Changes your agent makes appear instantlyRecommended for every developer
Coalesce platform skillsTeach your agent how to build Coalesce pipelines correctlyRecommended when your agent supports skills or plugins
Transform MCPRead and administrative access to your Coalesce organization: run history, job status, deployed environmentsRecommended for monitoring; the only piece read-only users need

CLI or MCP?

Install both for full coverage. They serve different halves of the workflow:

  • Use the CLI to build. Creating nodes, validating the workspace, planning, deploying, and running pipelines all happen through coa against your local files.
  • Use the MCP to observe. Checking whether last night's runs succeeded, diagnosing a failing job, or querying a deployed Environment with authority are what Transform MCP is good at.

A viewer or administrator who never edits pipelines can skip the CLI entirely and connect only the MCP.

Getting Set Up

The Local Development Setup Guide walks through the full installation: prerequisites, the CLI, the Desktop app, platform skills, authentication, and a smoke test that proves your agent can build.

What's Next?