cancel
Showing results forย 
Search instead forย 
Did you mean:ย 
Get Started Discussions
Start your journey with Databricks by joining discussions on getting started guides, tutorials, and introductory topics. Connect with beginners and experts alike to kickstart your Databricks experience.
cancel
Showing results forย 
Search instead forย 
Did you mean:ย 

Genie Code chat history monitoring of other users

Ramesh_01
New Contributor II

Hi community,

My organization has asked me to monitor how our team uses Genie Code (formerly Databricks Assistant). I'm a regular workspace user, not an admin, and so are my teammates. Asking each person to share their chats individually isn't practical, so I'd like to know what is supported.

1. Is there any supported, auditable way (UI, REST API, etc..) for a designated reviewer, with permissions granted by an admin, to see the prompts and responses of other users' Genie Code chats without each user sharing them? Or is chat history visible only to the user who created it by design?

2. Can a user export their own Genie Code chat history programmatically (for example with a PAT or OAuth token), or is the UI the only way?

Thanks!

2 ACCEPTED SOLUTIONS

Accepted Solutions

anuj_lathi
Databricks Employee
Databricks Employee

Short answer: As far as I know, there's no supported way for a delegated reviewer to read other users' Genie Code prompts and responses, and no public API to export your own chats. Don't build your monitoring plan around either. Genie Code chats are a per-user experience in the UI [1].

1. Reviewing other users' Genie Code chats

Don't confuse Genie Code with Genie Agents (formerly Genie Spaces). Agents have a review model, and Genie Code doesn't:

  • On a Genie Agent, users with CAN MANAGE can use the Monitor tab to see questions and responses. They can open full conversations that are set to Reviewable by agent managers. For Private conversations, they only see the prompts [2].
  • Admins can also query system.access.audit for Genie Agent events such as feedback and review requests [3]. Those queries filter on service_name = 'aibiGenie'.

None of that applies to Genie Code. It's a different surface with different controls, and those Monitor tab and aibiGenie audit queries only cover Genie Agent conversations. I wouldn't assume the audit log carries Genie Code prompt text. Have a workspace or account admin check what appears in system.access.audit for your workspace, and confirm with your Databricks account team if you need a firm answer for compliance purposes.

What does exist for oversight:

  • Adoption and impact tracking. The Genie Code docs have a "Measure Genie Code impact" section for tracking adoption across your organization [1]. That gives you usage-level visibility, not conversation content, and it's something an admin would set up or surface for you.
  • Data-access governance. Genie Code is governed by Unity Catalog permissions, so it can only touch data and operations the user already has rights to [1]. If your organization's real concern is "what is the team doing with data via Genie Code," the control point is UC grants plus audit logging, not chat review.
  • Cost and budget controls. Account admins can manage budgets and cost controls for Genie, since Genie Code is now billed pay-as-you-go with a per-user free allowance [1].

If the requirement really is "a reviewer reads prompts and responses," you have two options:

  1. Make it an explicit team practice that people share or paste chats into a review channel.
  2. Raise a feature request through your Databricks account team.

2. Exporting your own chat history programmatically

I'm not aware of a public REST API, SDK call, or CLI command that exports Genie Code chat history with a PAT or OAuth token. Today it's UI-only: chats persist as you move between pages, and you can reopen them there [1].

Don't mix this up with the Genie Agents API. That API (conversation and management endpoints) works on Genie Agent conversations [4]. It won't return your Genie Code threads.

What I'd do

  1. Ask an admin to review the Genie Code impact and usage reporting [1] and what shows up in the audit logs.
  2. If the requirement is content review, set that expectation through team process rather than tooling, or file a feature request.
  3. Don't build a scraper against internal UI endpoints. They aren't supported and can change without notice.

References

[1] Genie Code | Databricks on AWS โ€” https://docs.databricks.com/aws/en/genie-code/
[2] Test and monitor a Genie Agent | Databricks on AWS โ€” https://docs.databricks.com/aws/en/genie-agents/monitor
[3] Monitor Genie Agent usage with audit logs and alerts | Databricks on AWS โ€” https://docs.databricks.com/aws/en/genie-agents/audits-alerts
[4] Use the Genie Agents API | Databricks on Google Cloud โ€” https://docs.databricks.com/gcp/en/genie-agents/conversation-api

Anuj Lathi
Solutions Engineer @ Databricks

View solution in original post

rragi
Databricks Partner

Hi Ramesh,

You cannot export genie code(code assistant) chats but you can audit, govern and review responses on Genie Agent. 
The same applies for others agents as well but not for the chat assistant. https://docs.databricks.com/aws/en/genie-agents/monitor 

 

View solution in original post

2 REPLIES 2

anuj_lathi
Databricks Employee
Databricks Employee

Short answer: As far as I know, there's no supported way for a delegated reviewer to read other users' Genie Code prompts and responses, and no public API to export your own chats. Don't build your monitoring plan around either. Genie Code chats are a per-user experience in the UI [1].

1. Reviewing other users' Genie Code chats

Don't confuse Genie Code with Genie Agents (formerly Genie Spaces). Agents have a review model, and Genie Code doesn't:

  • On a Genie Agent, users with CAN MANAGE can use the Monitor tab to see questions and responses. They can open full conversations that are set to Reviewable by agent managers. For Private conversations, they only see the prompts [2].
  • Admins can also query system.access.audit for Genie Agent events such as feedback and review requests [3]. Those queries filter on service_name = 'aibiGenie'.

None of that applies to Genie Code. It's a different surface with different controls, and those Monitor tab and aibiGenie audit queries only cover Genie Agent conversations. I wouldn't assume the audit log carries Genie Code prompt text. Have a workspace or account admin check what appears in system.access.audit for your workspace, and confirm with your Databricks account team if you need a firm answer for compliance purposes.

What does exist for oversight:

  • Adoption and impact tracking. The Genie Code docs have a "Measure Genie Code impact" section for tracking adoption across your organization [1]. That gives you usage-level visibility, not conversation content, and it's something an admin would set up or surface for you.
  • Data-access governance. Genie Code is governed by Unity Catalog permissions, so it can only touch data and operations the user already has rights to [1]. If your organization's real concern is "what is the team doing with data via Genie Code," the control point is UC grants plus audit logging, not chat review.
  • Cost and budget controls. Account admins can manage budgets and cost controls for Genie, since Genie Code is now billed pay-as-you-go with a per-user free allowance [1].

If the requirement really is "a reviewer reads prompts and responses," you have two options:

  1. Make it an explicit team practice that people share or paste chats into a review channel.
  2. Raise a feature request through your Databricks account team.

2. Exporting your own chat history programmatically

I'm not aware of a public REST API, SDK call, or CLI command that exports Genie Code chat history with a PAT or OAuth token. Today it's UI-only: chats persist as you move between pages, and you can reopen them there [1].

Don't mix this up with the Genie Agents API. That API (conversation and management endpoints) works on Genie Agent conversations [4]. It won't return your Genie Code threads.

What I'd do

  1. Ask an admin to review the Genie Code impact and usage reporting [1] and what shows up in the audit logs.
  2. If the requirement is content review, set that expectation through team process rather than tooling, or file a feature request.
  3. Don't build a scraper against internal UI endpoints. They aren't supported and can change without notice.

References

[1] Genie Code | Databricks on AWS โ€” https://docs.databricks.com/aws/en/genie-code/
[2] Test and monitor a Genie Agent | Databricks on AWS โ€” https://docs.databricks.com/aws/en/genie-agents/monitor
[3] Monitor Genie Agent usage with audit logs and alerts | Databricks on AWS โ€” https://docs.databricks.com/aws/en/genie-agents/audits-alerts
[4] Use the Genie Agents API | Databricks on Google Cloud โ€” https://docs.databricks.com/gcp/en/genie-agents/conversation-api

Anuj Lathi
Solutions Engineer @ Databricks

rragi
Databricks Partner

Hi Ramesh,

You cannot export genie code(code assistant) chats but you can audit, govern and review responses on Genie Agent. 
The same applies for others agents as well but not for the chat assistant. https://docs.databricks.com/aws/en/genie-agents/monitor