Between September 22 and October 6, 2026, Databricks announced three command line tools, and their names say little about what they do. In short: the Agent Bricks CLI builds and deploys an agent you write in code. Genie Code CLI is a coding agent that works for you in the terminal. The Unity Gateway CLI, ug, is not an agent at all: it connects the coding agent you already use, such as Claude Code or Codex, to your company's Databricks setup. The first two need the older Databricks CLI installed.
Key facts
- Agent Bricks CLI (
agentbricks): Beta since September 29, 2026. For developers who ship their own agent. - Genie Code CLI (
genie): generally available since November 18, 2026, after a Beta from October 6. Databricks supplies the models and bills them as Genie usage. - Unity Gateway CLI (
ug): announced September 22, 2026. It sets up and launches other agents through Unity Gateway. - Databricks CLI (
databricks): the base tool. The first two need it installed and sign in with its saved profiles.
Everything below comes from the Databricks documentation as it read on November 21, 2026. We have not run these tools ourselves for this piece, so where we describe behavior, it is what the docs say, with the page named in the sources.
The four Databricks CLIs at a glance
| Tool | Command | What it is | Stage | Use it when |
|---|---|---|---|---|
| Agent Bricks CLI | agentbricks | A build and deploy tool for custom agents written in code | Beta | You are shipping your own agent to the Databricks agent runtime |
| Genie Code CLI | genie | A coding agent in your terminal, tuned for Databricks work | Generally available | You want an agent to do data and AI work for you, on models Databricks provides |
| Unity Gateway CLI | ug | A connector that configures and launches other coding agents | No Beta or Preview label on its pages | You already use Claude Code, Codex or another agent and want it governed by your workspace |
| Databricks CLI | databricks | The general tool that wraps the Databricks REST API | Generally available from version 1.0.0 | You script or automate the platform itself |
Agent Bricks CLI: build and deploy your own agent
The Agent Bricks CLI is for developers who write agent logic in code. The package is called databricks-agentbricks, but the command you type is agentbricks, which is why searches for "agentbricks cli" and "agent bricks cli" both lead here. It starts a project from a template, runs it on your machine, and deploys it to the Databricks agent runtime. Model access, memory, sessions, tools and MLflow tracing are set up for you.
The docs give the whole path from sign-in to a deployed agent in six commands:
databricks auth login --host "https://<your-workspace-url>" --profile my-workspace agentbricks login --profile my-workspace agentbricks init --framework langgraph my-agent cd my-agent agentbricks dev agentbricks deploy my-agent
What to know before you start:
- Two templates.
--frameworktakeslanggraphfor LangGraph oropenaifor the OpenAI Agents SDK. - One file decides what gets deployed.
agent.tomllists the tools, memory, sessions and tracing your agent depends on, andagentbricks deployreads it to create and connect them. - What you need installed. The Databricks CLI,
uv, and Python 3.11 or above for the projects it creates. Databricks recommends Python 3.12. - Local runs have no memory.
agentbricks devstarts a local server on port 8000 but does not connect to memory or session stores. Memory connects only after you deploy. - It can adopt an agent you already have.
agentbricks init --existingdoes not rewrite your code. It writes migration instructions that a coding agent such as Claude Code or Codex follows, andagentbricks doctorchecks the result. - No workspace switch is needed. The Beta notice says no workspace setting is required: you install the CLI and begin.
Gotcha: tools that act as the signed-in user. If an MCP tool is added with user authentication, the deployed agent needs a signed-in user's token on every request. Calls made with a plain bearer token, such as curl or a CI job, fail with MCP_USER_AUTH_REQUIRED. For an agent you call from code, the docs say to add the tool with --auth app.
This is not the tool for someone who wants an agent without writing code. The same docs page points those readers to Genie Agents in the Databricks UI. Our Agent Bricks CLI roadmap page tracks its status.
Genie Code CLI: a coding agent in your terminal
Genie Code CLI is the one that does work for you. It runs on your machine, reads and edits local files, runs commands, and reaches into Databricks under your own identity. It works like other terminal coding agents, with three differences the docs spell out: it comes set up for Databricks, it uses models that Databricks provides so you need no model subscription of your own, and whatever it does in the workspace is limited by your Unity Catalog permissions.
Install it with one command on Linux or macOS (Windows has a PowerShell equivalent), then start it in a project folder:
curl -fsSL https://github.com/databricks/genie-code-cli/releases/latest/download/install.sh | bash cd my-project genie
What to know before you start:
- Requirements. Unity Catalog on the workspace, partner-powered AI features enabled for both the account and the workspace, and Databricks CLI version 1.0.0 or above installed separately.
- Sign-in is OAuth only. It works with a Databricks CLI profile created by
databricks auth login. Personal access tokens and machine-to-machine OAuth are not supported. - Billing. It is billed like Genie Code in the workspace, on Genie pay-as-you-go pricing, and shares each user's free monthly Genie usage. Admins manage one Genie Code budget across the CLI and the workspace.
- It is a separate product from Genie Code in the workspace. It has its own tools, skills and instructions, does not share them with the workspace version, and does not yet have every workspace capability.
- Skills are local only. It loads skills from folders on your machine, such as
~/.genie/skills/, and not from your workspace or from Unity Catalog. - It can import your setup. If you use Claude Code or Cursor,
/importbrings your configuration across. Project rules go in anAGENTS.mdfile. - Regions. It is not available where Genie pay-as-you-go billing is not: Azure West Central US and Qatar Central, AWS Asia Pacific (Jakarta), and government regions.
One naming trap: the genie command is Genie Code CLI, while databricks genie is a group of commands inside the Databricks CLI, such as databricks genie ask. Genie Code CLI calls those commands to find data and answer questions. For how Genie Code relates to the other Genie products, see Genie One vs Genie Agents vs Genie Code; status is on our Genie Code CLI roadmap page.
Unity Gateway CLI (ug): connect the agent you already use
ug does not write code and has no model of its own. It signs you in, writes the connection settings for your coding agents, and launches them through Unity Gateway, applying your organization's configuration for models, MCP tools and skills. The agent keeps the interface you already know.
uv tool install git+https://github.com/databricks/unity-gateway ug configure ug claude
What to know before you start:
- Agents it launches.
ug claude,ug codex,ug gemini,ug opencode,ug copilotandug pi.ug configurealso sets up the Claude and ChatGPT desktop apps and GitHub Copilot in Visual Studio Code. - Cursor is a partial case. For Cursor,
ugsets up MCP tools only, and model requests from Cursor Agent use your Cursor account. - Tools, skills and spend.
ug mcp addandug skills addadd approved MCP servers and skills.ug usageshows your personal and shared spend limits in the terminal. - What you need. Python 3.12 or later,
uv, and the Workspace access or Consumer access entitlement. - Sign-in. OAuth by default, or a personal access token with
--use-pat, which Genie Code CLI does not accept. - It installs from GitHub. The install command points at a Git repository, not PyPI, which is worth telling whoever approves software on engineer laptops.
The name changed. When we wrote about skills in Unity Catalog on September 2, 2026, the docs called this tool ucode and installed it from a repository of that name. The current pages use ug and the unity-gateway repository, and we found no mention of ucode on them. If an older tutorial tells you to run ucode configure, the command today is ug configure.
Status and history are on our Unity Gateway CLI roadmap page.
Which one do you need?
| Your job | Start with |
|---|---|
| You are building an agent as a product or internal service | Agent Bricks CLI. It is built for custom agents written in code. |
| You want help building pipelines, models or apps, and have no coding agent set up | Genie Code CLI. Nothing to configure beyond the install and sign-in. |
| You already use Claude Code, Codex, Copilot or Gemini CLI | ug. You keep your agent and gain governed models, tools and budgets. |
| You run a platform team | ug for governing the agents people bring, plus the Genie budget settings for Genie Code CLI. |
| You write deployment scripts or CI jobs | The Databricks CLI. None of the three replaces it. |
On the choice between Genie Code CLI and your own agent through ug, the Genie Code CLI page gives a direct answer. Use another agent through ug if you have already set it up with Databricks skills and managed MCP servers such as Genie One. Use Genie Code CLI if you have not, or if you want the extra tuning Databricks applies to make the agent work better with its platform.
How they fit together
The three are layers, not rivals, and a team can end up using all of them:
- Unity Gateway sits under the other two. Genie Code CLI usage is tracked through Unity Gateway, and
ugexists to put outside agents behind the same gateway. - The Agent Bricks CLI expects a coding agent beside it. Its migration path hands instructions to Claude Code or Codex, which you might be running through
ug. - Genie One connects to both. Genie Code CLI can add the Genie One MCP server with
genie mcp add, and an Agent Bricks project can add it as a tool withagentbricks tools add mcp system.ai.genie_one_mcp. - The Databricks CLI is the base. Agent Bricks CLI and Genie Code CLI list it as a requirement, and its saved profiles are how you sign in to them.
Sources
- Agent Bricks CLI, Databricks docs, last updated November 17, 2026
- Genie Code CLI, Databricks docs, last updated November 16, 2026
- Get started with coding agents, Databricks docs, last updated November 17, 2026
- ug CLI reference, Databricks docs, last updated November 20, 2026
- Supported coding and work agents, Databricks docs, last updated November 17, 2026
- Databricks CLI, Databricks docs, last updated October 20, 2026
- Release notes: Agent Bricks CLI is in Beta (September 29, 2026), Genie Code CLI is in Beta (October 6, 2026), Genie Code CLI is now generally available (November 18, 2026), and Connect coding agents to Unity Gateway with the ug CLI (September 22, 2026)
Checked against the Databricks documentation on November 21, 2026. The Agent Bricks CLI is in Beta and all three tools are changing quickly, so verify commands against the docs before you build on them. brickster.ai is an independent community project and is not affiliated with Databricks.



