Genie Code CLI

Genie Code CLI is a Beta coding agent from Databricks that runs locally in your terminal and is tuned for data and AI work on Databricks. It reads and edits local files, runs commands, and uses the Databricks CLI to reach into your workspace. It was announced in the October 6, 2026 release notes.

What It Is

Genie Code CLI works like other terminal coding agents but comes preconfigured for Databricks. It includes Databricks agent skills and uses the Databricks CLI, including databricks genie commands such as databricks genie ask, to discover and query Unity Catalog data. You start it with genie from a project directory and select a Databricks CLI profile for authentication. Genie Code CLI runs under your own Databricks identity, so workspace operations are limited by your permissions.

Example tasks in the docs include cleaning CSVs and loading them into Unity Catalog, extending a pipeline, training and serving a model, and building and deploying a Databricks App.

Setup and Requirements

The CLI needs a Databricks workspace in a Unity Gateway supported region, Unity Catalog enabled, and the Databricks CLI version 1.0.0 or above installed separately. It is installed with a curl script on Linux or macOS or a PowerShell script on Windows, and you can confirm the install with genie --version. It can be started from the integrated terminal of IDEs such as Visual Studio Code, Cursor, or IntelliJ IDEA. genie resume and /resume bring back earlier conversations, and genie update updates the tool.

Skills, MCP, and Project Instructions

You can add custom skills in the standard Agent Skills format under ~/.genie/skills/, or share them with a team in .agents/skills/ in a repository. genie mcp add connects remote or local MCP servers, including managed Databricks MCP servers such as Genie One. An AGENTS.md file in the project root supplies project-wide instructions. Running /import brings in configuration from Claude Code or Cursor.

Governance and Beta Limitations

Model access is provided and governed through Unity Gateway, so no separate model subscription is needed, and usage appears in Unity Gateway usage and budgets. During Beta, model selection is managed by Databricks (GPT 5.6 is documented) and cannot be chosen. Documented limitations include OAuth M2M authentication only, no SSH tunnel sessions, skills loaded only from local folders, and MCP calls failing with a 401 error once an access token expires. It is a separate experience from Genie Code in the workspace and does not yet include all of its capabilities.

Billing

CLI usage is billed at the same rate as accessing its model directly through Unity Gateway. Charges appear in Unity Gateway usage, and Unity Gateway budgets apply. Genie pricing and Genie budgets do not apply to this CLI. Running Databricks jobs, serving models or using other workspace resources can incur their own charges.