Trouble while Maintaining Databricks Genie Agents (VS code extension available?)

[DB1To3]

Is there an update to the databricks extension for VS code that will allow me to work with "genie agents" (aka spaces)?

I'm having a really hard time with the editing of these agents in the Azure workspace environment.  There are lots of things missing, like copy/paste from one agent to another.  Developers will often need to fork off of one piece of software to play with tentative changes, and then re-introduce their changes to the original copy at a later time (if everything worked as expected).  But simple copy/paste operations are not available in here.  Nor is there any text-based representation of "sources" or "examples" (FKs) that can be edited.  Even if the UX provided a back-door way to edit json or xml would be fine.

If there is some sort of preview channel for the VS code extension, please let me know!

More ranting:  After just a few attempts to edit agents, it triggered a new kind of migraine so I'll be heading home for the day.  Consider the following list of "sources" (image), notice that the tables were added over time and they aren't listed alphabetically.  There appears to be no way in this portal to re-sequence them after the fact.  

It would be nice if I can escape the browser, and do some of this in VS code.

[peter89jean]

Maintaining Databricks Genie Agents (formerly Genie Spaces) entirely through the browser portal can be frustrating, especially given the lack of robust text-based configuration files, straightforward mass-editing features (like bulk copy/pasting sources), or native table sorting within the workspace UI.

Currently, Databricks handles Genie space/agent metadata primarily through its managed workspace interface rather than exposing direct configuration files inside the VS Code extension. However, you can streamline your workflow and mitigate these maintenance bottlenecks by leveraging programmatic alternatives or utilizing Genie Code workflows.

[VladTamas]

Hi! There's no Genie editor in the Databricks VS Code extension yet, but you can get most of what you're asking for by managing the agent as code with Declarative Automation Bundles (DABs, formerly Databricks Asset Bundles). It works the same on Azure.

1. Export the agent to files (Databricks CLI 1.3.0 or newer; the command still says "genie-space" even though the product is now Genie agents):
databricks bundle generate genie-space --existing-id <AGENT_ID> --key my_agent

This writes src/my_agent.geniespace.json (the full agent: tables, instructions, example SQL) and resources/my_agent.genie_space.yml. You can edit, diff and copy/paste between agents in VS Code, and keep it all in Git.

2. Fork, test, merge back: use Clone from the agent's ⋮ menu to make a test copy (it copies tables, instructions, example SQL and functions). Once you're happy with the changes, move them into the original agent's JSON and run databricks bundle deploy. Pull request reviews come for free once it's in Git.

3. Pull UI edits back into the files:
databricks bundle generate genie-space --resource my_agent --force
(add --watch to keep syncing while you edit in the UI)

4. Sorting: in the JSON format, the tables are always sorted alphabetically, because the format requires it.

Two things to watch: Genie resources only deploy with the direct engine (the default for new bundles; older bundles need engine: direct in databricks.yml), and you'll need a recent CLI.

Full walkthrough with an example repo: community.databricks.com/technical-blog/managing-databricks-genie-agents-as-code-with-databricks/ba-...
Resource reference: learn.microsoft.com/en-us/azure/databricks/dev-tools/bundles/resources

Hope this helps!

[anuj_lathi]

Short answer: the Databricks extension for VS Code doesn't have a Genie space editor, and I'm not aware of one on a roadmap I can point you to. Genie space authoring (sources, instructions, examples) lives in the workspace UI, so you can't currently move that editing into VS Code.

What the extension does cover

Its scope is bundles (define, deploy, run), running Python files and notebooks on clusters or serverless, Databricks Connect debugging, and syncing local code to the workspace [1]. That's code and job/pipeline development, not editing workspace-native objects like Genie spaces. I can't confirm a preview channel with Genie support. If you want to check, look at the extension's page in the VS Code Marketplace for a "Switch to Pre-Release Version" option, but I wouldn't expect it to change this.

Workarounds for the fork-and-merge workflow

1. Keep the "text" of the space in source control, outside the space. Most of what you're trying to copy is instructions, example SQL, and table/column descriptions. Keep the instruction text and example question/SQL pairs in a repo as Markdown or SQL files, and paste from there. It's clunky, but it gives you diffs and history, which the UI doesn't.
2. Push metadata into Unity Catalog instead of the space. Table and column comments and descriptions live with the tables, so every space that uses those tables inherits the improvements. You can manage them as SQL in VS Code and run them through the extension [1]. Less per-space config also means less to copy between spaces.
3. Use the Genie REST APIs for a programmatic, text-based view. Look in the API reference for your workspace to see what's available for retrieving and creating spaces. If your workspace exposes a way to export and import a space definition, that gives you the JSON representation you're asking for, and you can diff it and script the copy from a dev space to a prod one. I'd verify the exact endpoints and payload shape on your platform before building on them, since this area has been changing quickly.
4. Clone rather than copy/paste. For forking, duplicate the space in the UI (if your workspace offers it), experiment in the clone, and re-apply the changes you want to keep. Pair that with the repo-based approach from step 1 so the "changes to re-introduce" are a diff rather than something you remember.

Source ordering

I don't know of any way to re-sequence the sources list in the UI. It appears to be in the order the tables were added. If the order matters to you, the only option today is removing and re-adding tables, which is painful with 12 of them. Ordering is unlikely to affect Genie's answer quality, so I'd treat it as cosmetic.

Next step

The requests you describe (copy between spaces, a JSON/text view, reordering sources, IDE editing) are all reasonable feature requests. File them through your Databricks account team or the Ideas portal, since that's the channel that gets them in front of the Genie product team. In the meantime, I'd go with the repo-plus-API approach in step 3 as the closest thing to "escaping the browser."

References

[1] Databricks IDE extension | Databricks on AWS — https://docs.databricks.com/aws/en/dev-tools/vscode-ext