The Unity Gateway Cli

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: The Unity Gateway Cli · Channel: Databricks · 04:00 · Watch on YouTube

Topics: AI agents, Data, databases and embeddings, AI developer tools, AI governance and regulation, Model inference and optimization

Summary

The video demonstrates how to configure and use the Unity Gateway CLI to manage governed coding agents in Databricks. The presenter checks usage and starts Codex, then uses Claude Code to ask a question that is routed to a more capable model through smart routing. She verifies the answer and trace in the Databricks workspace, and shows how administrators can configure agents, models, MCP servers, skills, and budget-based defaults. The demo ends by adding a model through a JSON configuration file and highlighting how the CLI helps teams centrally manage and govern AI tools.

Key points

Unity Gateway is configured from a terminal with the UG configure command and one-time Databricks workspace authorization.
The CLI configuration displays the workspace, available coding agents, MCP servers, and skills.
Users can check their usage and budget before starting a coding-agent session.
Codex provides a choice of available GPT models and reasoning levels.
Unity Gateway smart routing selects a more performant model for complex, open-ended questions.
The presenter checks the Claude Code response and its trace in the Databricks workspace.
Administrators can configure default agents, models, MCP servers, skills, and smart-routing behavior.
The configuration can be edited as JSON to add models and define budget-based defaults.
Unity Gateway CLI lets administrators centrally manage and govern tools for their teams.

Chapters

Start	Chapter	Summary
00:03	Configuring Unity Gateway and checking usage	The presenter configures Unity Gateway, reviews the available agents and tools, checks her budget, and explores model choices in Codex.
01:10	Smart routing with Claude Code	She asks Claude Code how many stores Boba Bricks has, and Unity Gateway routes the open-ended question to Claude Opus 5.
02:15	Admin settings and agent configuration	The demo covers agent defaults, available models, smart routing, tracing, MCP servers, skills, and budget-based configuration.
03:25	Editing configuration and governing team tools	The presenter adds a model by editing the JSON configuration and describes how administrators can manage and distribute AI tools for their teams.

Quotes

First, I’m going to configure Unity Gateway through my terminal using UG configure. — 00:03

it’s immediately kicked off Unity Gateway smart routing. — 01:10

all that’s stored in my unified traces view. — 02:50

The Unity Gateway CLI is available now, and I can’t wait for you to check it out. — 04:00

Concepts and entities

Unity Gateway CLI, Databricks, Claude Code, Codex, smart routing, MCP servers, Boba Bricks

Full transcript (original language)

00:03 Hi everyone, Amber here from Databricks Tech Marketing and I’m here to show you the latest capabilities for managing governed coding agents with the Unity Gateway CLI. First, I’m going to configure Unity Gateway through my terminal using UG configure. This will require a one-time authorization into my Databricks workspace. And then after that authentication is complete, Unity Gateway is connected. So in the configuration, we can see the workspace that we’re working from, the coding agents we have available, so Claude Code and Codex, as well as MCP servers and skills. Next, I want to be

00:38 able to check my usage before I start a session and it looks like I’ve only spent $41 of my $5,000 budget. So I’m good to start Codex. In Codex, I want to see which GPT models are available for my use. So currently, the default is GPT 5.6 Soul, but I have a lot of models available, so I can even choose a model like 5.6 Luna. I’m able to select the reasoning level. And actually, I want to switch to our default coding agent, Claude Code.

01:10 So I’m going to exit that session and then type UG. Now, I work with a company called Boba Bricks and so I want to ask Claude, “How many stores does Boba Bricks have?” And we can see that it’s immediately kicked off Unity Gateway smart routing. This means for more complex tasks like open-ended questions, like the question I just asked, it’s going to use a more performant model, so it’s choosing Claude Opus 5 to run this query through.

01:44 And here we can see the output that Boba Bricks has 14 stores and a regional breakdown. And now when I exit the Claude Code session, I once again can see my budget, so how much I used and how much I still have remaining. Now going into my Databricks workspace into my Unity Gateway page, I’m going to click on traces and look at the latest Claude code trace just to see that the output is consistent from the terminal. And here we do see that Bob bricks has 14 stores and the source

02:15 showing that my traces are configuring correctly and being stored. Now going to my agent configuration, and this is where I as an admin could set up all the defaults. So, all my developers and engineers can use coding agents without having to set everything up themselves. Starting with the default coding agent you saw when I typed UG, we got Claude code, and then we have various models available. And because I’ve also enabled smart routing, even though we have Sonnet 5 as the default, Opus 5 answered

02:50 my question because it was the right model to answer that question. If it was a lot more straightforward of a question, we could have used a lower cost model like Kimmy. Now going to Codex, we can see the models that are available through Codex. And again, smart routing and tracing was fully enabled for both coding agents, and all that’s stored in my unified traces view. We have the MCP servers that are available as well as skills, and then of course smart defaults. So, I’ve only used a small amount of my budget, but if I got to say 80% of my budget, it would switch to using Codex as the default

03:25 with a Kimmy model. And if I want to edit the configuration, whether that’s the default agent, the MCP servers, the skills, I can go and directly edit the JSON file. Recently, GPT-6 Astra came out, so I want to add that to my model services. I can save the JSON. And then now we can see that model is available in Codex. And this isn’t just for me. As an admin, this is how I can manage, distribute, and centrally govern

04:00 all these AI tools for my team. So they can easily install and get started without having to install coding agents and tools from scratch. The Unity Gateway CLI is available now, and I can’t wait for you to check it out. Thank you.