All Generative AI posts

Re: How to retrieve Genie's narrative text summary (final_summary) via API?

no2 — Mon, 29 Jun 2026 14:54:23 GMT

I understand that this is a fast moving world right now, but the limitations of the Genie API are a constant source of frustration for my company.

I find it strange that your response says "as there appears to be a good amount of demand for it."

You just need to assume that any feature released in the UI is also in-demand for the API. Ideally the features should be released at the same time.

MCP on Databricks

sagarpgowda7777 — Sun, 28 Jun 2026 16:51:18 GMT

Here’s What Nobody Tells You.
A hands-on look at Genie MCP and DBSQL MCP — what works, what doesn’t, and when to skip MCP entirely.
Let me start with something most MCP content skips. MCP servers don’t just expose tools. They expose three things — tools, prompts, and resources. If you only think about tools, you’ll miss why some things break in ways that aren’t obvious.
I validated this on two Databricks MCP servers: Genie and DBSQL. Here’s what I actually found.
Genie MCP — The Idea Is Right. The Execution Needs Time.
I went in wanting one thing: route queries across multiple Genie spaces without writing routing code for each one. That part does work. But everything around it is still rough.
Query Space
Genie MCP exposes descriptions of your Genie spaces as resources so the LLM can decide which space to query. In theory, smart. In practice, the descriptions are too generic. If you have three Genie spaces and their descriptions are all high-level overviews, the model doesn’t have enough signal to route correctly. You get misrouting or arbitrary selection. This will improve as the feature matures — but right now, you need to write much more specific descriptions manually to make routing reliable.
Poll Responses
Genie is async. You fire a query, get a conversation ID, then poll until the response is ready. The MCP server exposes this flow, but the polling configuration is still on you. You still need retry logic, timeout handling, error cases. MCP wraps the API call — it doesn’t solve the async complexity underneath.
Statelessness — This Is the Real Gap
Every Genie MCP call starts a new session. No state carried across tool invocations. If your agent needs to follow up based on a previous answer, that context is gone. For any conversational agent, this is a hard limitation. The direct Genie Conversation API handles this better because you manage the conversation ID yourself — more code, but you own the state.

Area:What’s missing today

Query Space :Descriptions too generic for reliable LLM routing

Poll Responses :Polling still needs to be configured manually — MCP doesn’t abstract that away

Statelessness Each call starts a new session — no conversation continuity across turns

Best workaround Custom MCP server wrapping the Genie API — routing benefits, full control over state and polling

Where Genie MCP Actually Wins

Multiple Genie spaces. If you’re orchestrating across three or four spaces, the multi-server MCP pattern lets the LLM route from server config. Writing that same thing over raw API means a tool function per space, explicit routing conditionals, and ongoing maintenance. MCP removes that layer.

There’s also a third option worth considering: a custom MCP server wrapping the Genie API. You get the routing benefit of MCP without giving up control over state and polling. That’s the direction I’ve been moving toward.

DBSQL MCP — This One’s Ready

Different story here. DBSQL MCP feels production-ready.

Three tools: execute_sql, execute_sql_read_only, and poll_sql_result. The read-only tool is what I’d default to for agentic workflows — full query capability with no risk of an agent running an accidental write.

Tool Use it when

execute_sql: You need full read/write in a controlled pipeline

execute_sql_read_only :Default for agents, VS Code/Copilot integrations, anything exploratory

poll_sql_result :Query is async or long-running — fetch results after completion

Unity Catalog Integration — This Is What Makes It Work

The MCP server respects your UC permissions exactly. You see the tables you have access to. Masked columns stay masked. Encryption policies hold. It doesn’t give the LLM more access than you’d have in the Databricks UI yourself. That governance consistency matters when you’re connecting external tools to your lakehouse.

Authentication is PAT or SSO. The server’s access boundary is your workspace access. Nothing more.

Practically, this means you can query your lakehouse directly from VS Code or GitHub Copilot while writing code, feed governed table data into an agentic reasoning loop, or let external tools run SQL without needing direct cluster access — all while staying within whatever access policies you’ve already set in Unity Catalog.

The Bottom Line

DBSQL MCP is worth using today. Genie MCP is worth watching, but statelessness, routing quality, and the polling gap mean the direct API is still the better call if you need reliability and control.

The exception is multi-space routing. If that’s your main problem, Genie MCP saves real work. And if you need both — routing and control — build a custom MCP layer over the API.

The ecosystem is moving fast. Some of what I’ve described here may already look different by the time you read this. But this is what I found when I actually tested it.

Test it yourself. Don’t build on assumptions.

#MCP #Databricks #AIEngineering #AgenticAI #DataEngineering #LangGraph

Genie not able to list MCP prompts

sandyveeru — Fri, 26 Jun 2026 13:29:33 GMT

Hi community,

I'm building a custom MCP proxy server (hosted as a Databricks App) that forwards requests to an upstream MCP server. The upstream server exposes both **tools** and **prompts** as per the MCP specification.

I can confirm that Genie Code successfully discovers and calls tools via `tools/list` and `tools/call`. However, I'm unable to find any documentation or confirmation on whether Genie Code also supports the MCP **prompts** primitive — specifically:

- Does Genie Code ever call `prompts/list` on a connected MCP server?
- Does Genie Code ever call `prompts/get` to invoke a specific prompt?
- If prompts are supported, how does Genie Code surface or use them in Agent mode?

The [MCP documentation on Databricks](https://docs.databricks.com/aws/en/generative-ai/mcp/) describes MCP as a standard for *"tools, resources, prompts, and other contextual information"*, but all concrete examples and limits (e.g. the 20-item cap) only reference tools. There is no explicit mention of prompt discovery or invocation in Genie Code.

**My setup:**
- Custom MCP server deployed as a Databricks App
- FastMCP with `stateless_http=True`
- Server correctly responds to `prompts/list` and `prompts/get` (verified via MCP Inspector)
- Genie Code connects successfully and uses tools, but no evidence of `prompts/list` being called

Has anyone confirmed whether MCP prompts are supported, planned, or intentionally out of scope for Genie Code? Any official reference or firsthand experience would be appreciated.

Thanks!

Re: Agentbricks Knowledge Assistant - Facing Issue with Agent Creation and Source Visibility

Ajay-Pandey — Fri, 26 Jun 2026 09:45:18 GMT

Thanks @Ashwin_DSA

I have already tried the provided solution, but I am still facing the error. I have also raised a support request with Azure Databricks; however, it has been going in circles for the past month without resolution.

Re: AI Extract is disabled?

Ashwin_DSA — Thu, 25 Jun 2026 19:28:19 GMT

Hi @suranga,

I have checked this internally. What you're seeing is most likely a workspace availability issue rather than a problem with JPEG files in a Unity Catalog Volume. The public Information Extraction documentation lists a few workspace prerequisites for the UI, including serverless compute, Unity Catalog, access to a serverless usage policy with a nonzero budget, and region support for the feature.

On the file side, JPEGs in a Volume are a supported path. The public ai_parse_document documentation explicitly says files in a Unity Catalog Volume can be read with the binaryFile reader, and it lists JPG/JPEG as supported formats.

One subtle point is that ai_extract being GA does not automatically mean the no-code Information Extraction experience is available in every Azure workspace and region. The Azure release notes say ai_extract is GA, but the Information Extraction docs still note limited regional availability and point to the feature availability matrix.

It’s also worth noting that AI Extract does not take raw files directly. For document and image workflows, the public docs show it being used downstream of ai_parse_document, so the expected pattern is to parse first, then extract.

So the checks I'd suggest are:

Confirm the workspace region against the public feature availability page
Confirm serverless compute is enabled
Confirm Unity Catalog is enabled
Confirm the workspace has access to a serverless usage policy with a nonzero budget

If those prerequisites are already satisfied and the UI still shows "contact sales," I'd treat that as a workspace access or availability issue for the Information Extraction UI, not as a JPEG-in-Volume support issue.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.

Re: Accessing Document Presented in Demo in Get Started with AI Agents on Databricks Course

3R1 — Thu, 25 Jun 2026 19:04:55 GMT

Hi Ashwin,

Thank you for the follow up.

I have a hard time to find the content tab and will submit a ticket.

Fransis

Re: Accessing Document Presented in Demo in Get Started with AI Agents on Databricks Course

Ashwin_DSA — Thu, 25 Jun 2026 18:49:55 GMT

Hi @3R1,

Thank you for flagging this!

The document shown in the Demo section of the Get Started with AI Agents on Databricks course is a Databricks Notebook. You should be able to access it by following the steps below:

Log in to Databricks Academy at customer-academy.databricks.com
Navigate to the course session
Click the Content tab
Look for the Demo - Get Started with AI Agent on Databricks notebook link

If the link is not visible in the Content tab, please check the Resources section within the course module.

If you are already in the course and still cannot locate the document, please raise a Training Support ticket and the training team will help with this.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.

AI Extract is disabled?

suranga — Thu, 25 Jun 2026 16:09:18 GMT

Why is AI Extract is not loading up? im on azure databricks and i have JPEG images in a Databricks Volume. Its on GA but why is it asking to contact sales for this?

Accessing Document Presented in Demo in Get Started with AI Agents on Databricks Course

3R1 — Thu, 25 Jun 2026 14:54:12 GMT

I'm taking Get Started with AI Agents on Databricks Course and cannot find the document that is shown in the Demo section.

I wonder whether the Demo is for a show only or the learner can follow along. If the latter, where can I find it?

Course Link: https://customer-academy.databricks.com/learn/courses/4464/get-started-with-ai-agents-on-databricks?hash=da4cd4915b4bbd06eed1c89022d28c0bb70ac589&generated_by=1000317

Thanks,

Fransis

Re: Agentbricks Knowledge Assistant - Facing Issue with Agent Creation and Source Visibility

Ashwin_DSA — Thu, 25 Jun 2026 10:41:48 GMT

Hi @Ajay-Pandey,

From the snapshots, I would treat this as a possible UI/control-plane issue, especially if the source data is still visible through the API but not in the Knowledge Assistant UI.

A few things I would suggest checking first:

Confirm the workspace meets the basic Knowledge Assistant requirements... such as ... Serverless compute, Unity Catalog, Model Serving access, a nonzero serverless usage policy budget, and a supported region (the public docs list these under Knowledge Assistant requirements). Docs
Double-check the source format and file limits. For volume-based sources, the supported file types are txt, pdf, md, ppt/pptx, and doc/docx. The docs also note that files larger than 50 MB are skipped, and that PDF/DOC/DOCX/PPT/PPTX files with more than 500 pages are skipped as well. Docs
If the source is an AI Search index instead of files, verify that the index uses a supported embedding model. The current public docs list databricks-gte-large-en, databricks-bge-large-en, and databricks-qwen3-embedding-0-6b. Docs
If this is a file-based source, try a manual Sync. The docs mention that file-based knowledge sources need to be synced after updates, while index-based sources update automatically. Docs
Give the agent enough time if it was recently created or updated. The public docs note that agent creation and source sync can take up to a few hours, and once the agent is ready, the right-side panel should display the synced knowledge sources. Docs
Test the agent from the Build tab or from Playground and see whether it returns citations at all. The docs also mention using "View sources" to inspect the references the assistant is citing. Docs

If the API still returns the Knowledge Assistant and its knowledge sources, but the UI shows an empty Sources pane or displays "Could not retrieve the agent," I’d recommend opening a Databricks support case. That usually points more to a mismatch between the backend and the UI than to unsupported source files.

For the support ticket, it would help to include:

Workspace URL, cloud, and region
The Knowledge Assistant name and ID if available
The exact error text
Screenshots of the empty Sources UI and the retrieval error
Whether the problem affects only one assistant or also newly created assistants
The source type being used (volume, table, or AI Search index)
Sample file types and approximate file sizes/page counts
Whether manual Sync was attempted
API output from the Knowledge Assistants endpoints, especially the assistant state and the list of knowledge sources

The public API reference for Knowledge Assistants is here if you want to compare UI behaviour with backend state: API reference

Hope this helps.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.

Agentbricks Knowledge Assistant - Facing Issue with Agent Creation and Source Visibility

Ajay-Pandey — Thu, 25 Jun 2026 04:43:49 GMT

Hi Team,

I am facing an issue while creating a Knowledge Assistant in the Agent section. It was working fine earlier, and I had uploaded multiple PDF files (70K). However, it has now started behaving unexpectedly.

I am not able to see any source data in the UI, although it is visible via the API.
The agent is also not returning answers based on the source data.
When I try to create a new agent, I receive the following error:

"Could not retrieve the agent.
There was an error retrieving the agent. If the problem persists, please contact support."

Re: Genie space: Delta Tables or Metric View

balajij8 — Wed, 24 Jun 2026 03:03:21 GMT

Adding both the raw Delta table and the Metric View generally causes confusions to the Space. Genie relies entirely on the schemas, descriptions and metadata you provide to translate queries into optimized SQL. If you expose both objects containing the same underlying data it introduces evere semantic ambiguities (2 path dilemma) where Genie has to guess whether to write an on the run un-governed calculation against the raw columns or leverage the defined semantic logic in metric view. It can drastically increase overhead and heavily opens the door for errors and inefficiencies.

You can hide the raw and silver layer tables from the Genie Space configuration entirely (after testing via raw tables intially) and expose the curated Metric View later. You can establish a strict single source of truth by defining the dimensions, relationships and business measures (total slippage) inside a Metric View. You can add the tables to the space if you have separate, completely non overlapping dimension tables that aren't integrated into that Metric View. Just ensure that there is zero structural or logical overlap with the primary fact data.

You can keep the model focused ideally aiming for fewer than 15 objects per space to maximize efficiency. Restricting Genie's scope to just your governed Metric Views ensures that the underlying model doesn't get distracted by redundant schemas or competing definitions. It guarantees significantly higher accuracy, lower consumption and completely reliable deterministic generation for users

Re: Genie space: Delta Tables or Metric View

Andrewww — Tue, 23 Jun 2026 20:06:11 GMT

Thanks for the reply! Very useful.

If i add both the metric view and the raw table will this confuse the genie space? Or is it ok?

Re: Genie space: Delta Tables or Metric View

anshul2528 — Tue, 23 Jun 2026 19:52:30 GMT

Hi @Andrewww, it is fantastic to see you exploring Databricks Genie Spaces! Since you are currently in the initial experimentation phase and working with a streamlined sales dataset, starting by pointing Genie directly to your raw table is completely fine for a proof of concept.

However, as your data grows and you add more details, pointing Genie directly to raw data will cause friction. Think of Genie as a super smart intern. If you give it a messy, raw table, it will waste a lot of time guessing what the columns mean and might make mistakes.

Leaving the raw setup as-is simply forces Genie to expend extra compute resources and processing time on basic logic:

Extra Logic Processing: If a user asks for "total sales," Genie has to write more complex, multi-step SQL queries on the fly to deduce whether to sum the raw amount column or filter out cancelled orders first.
Slow Answers: Querying raw rows forces Genie to scan every single record for every question, meaning your users wait longer for answers.
Complex Join Computations: As you add more tables later, Genie will have to dynamically compute table relationships and keys for every session, which adds computational overhead compared to reading a pre-governed view.

Creating a clean, pre-modeled metric view changes everything because it gives Genie a clear map to follow. You can lock your business math such as excluding cancelled orders, directly inside the view. Genie just reads it, ensuring everyone gets the exact same answers with much better performance.

To scale this smoothly, think of your data as a production line that refines raw data into high-quality answers:

Bronze Layer: This holds your raw, unedited sales entries, which are great for storage but too noisy for Genie.
Silver Layer: This cleans up your records, strips out processing errors, and standardizes dates.
Gold Layer: This is your highly polished semantic view built on top of that clean data, holding your official business metrics. This is the absolute best place to point Genie.

To get started with this setup, the official documentation recommends keeping your space focused on one small business topic at first. You should also give Genie a benchmarking cheat sheet of 5 to 10 example questions your team will actually ask to track accuracy.

Your biggest quick win right now is configuring your metadata to speak plain English:

Use Clear Synonyms: Map business terms or shorthand like "Revenue" or "Spend" directly to your table's columns.
Add Field Descriptions: Write simple, clear comments on your columns in Databricks. Genie reads these comments natively, which instantly stops the guessing game and ensures highly accurate answers.

For a deeper dive into these optimization strategies, you can read the full Databricks AI/BI Genie Best Practices Guide.

Happy experimenting, and enjoy building out your space!

Genie space: Delta Tables or Metric View

Andrewww — Tue, 23 Jun 2026 17:24:38 GMT

I have started experimenting with Genie Spaces.

I have a delta table with sales information. It has a date column, customer, product and amount bought.

My question is, is it better to point the genie space to the raw table or create a metric view with all the possible measures someone might ask? Considering this will also be enriched in the future with more data.

Re: Genie spaces

bala_sai — Tue, 23 Jun 2026 15:05:53 GMT

Hey @prasuanu1222, For 10,000 daily users and 100+ requests per minute, plain iframe embedding is probably not enough by itself.

Databricks has workspace level throughput limits for Genie. For iframe/UI access, the documented limit is around 20 questions per minute per workspace across all Genie Spaces. For the Genie Conversation API free tier, it is around 5 questions per minute per workspace. So if the application is expected to receive 100+ requests per minute, that will go beyond the default limits.

For this scale, you should look at higher throughput options with Databricks. From the application side, you should also avoid sending every user question directly to Genie. At 10,000 daily users, many questions may be repeated or follow common patterns. A better design would be to put a controlled backend layer in front of Genie to handle caching, throttling, retry/backoff, and routing.

For example, common questions can be answered from cache, predefined SQL, or dashboards. Only new or truly ad hoc questions should go to Genie.

So for this use case, you should consider the following:

* Use iframe only if actual peak usage stays within the supported limits.
* For 100+ requests per minute, discuss higher throughput options with Databricks.
* Add an application/backend layer for caching, queueing, throttling, and routing.
* Use dashboards or predefined queries for repeated questions.
* Use Genie mainly for conversational or ad hoc analytics.
* If a custom UX is required, use the Conversation API with proper capacity planning instead of plain iframe alone.

Re: Genie spaces

prasuanu1222 — Tue, 23 Jun 2026 14:38:58 GMT

Thanks for answering the question one more clarification is we want to embeded this as iframe in a external application and this application is used by 10000 users daily which will go off the limits on the no of requests and the genie space will have more than 100 requests for a minute so how to handle that?

Re: Genie spaces

bala_sai — Mon, 22 Jun 2026 15:40:53 GMT

Genie Spaces do not expose a single fixed model choice for standard usage it uses a compound AI system.

For cost, I would split it into two areas:

The first one is the SQL warehouse cost. Genie still has to run SQL against the warehouse attached to the space, so this depends on the warehouse type/size, query runtime, concurrency, and how many questions users are asking. This is probably the easier part to estimate once we have some query history from a pilot.

The second one is Genie or LLM usage cost, this depends more on the number of users, how many questions they ask, and whether the usage is happening through the UI/embed experience or through the API.

For embedding, API is not necessarily the only path. If the requirement is just to let users access an existing Genie Space from another internal app, iframe embedding is probably the simpler option. It avoids building a custom API flow for every interaction.

A couple of things still matter though like users need access to the Genie Space and the underlying data, and the workspace admin has to allow that application/site for embedding. Also users can interact with the embedded Genie Space, but they cannot edit the space configuration from there.

If the requirement is a fully custom experience across many Genie Spaces, then the Conversation API may still be needed. In that case it is better to avoid a direct one to one integration for every space. A better design would be to have a common backend layer that routes requests to the right Genie Space, with caching, queueing/retry logic, and maybe fewer curated domain level spaces instead of too many small spaces.

Re: Genie spaces

Ashwin_DSA — Mon, 22 Jun 2026 15:19:14 GMT

Hi @prasuanu1222,

For the first question, it has been answered before on this forum. Genie is more of a managed service. You can find details about the underlying models on this page. This can change as Databricks switches to the best model available at the time.

From a cost perspective, Genie costs have two parts. LLM usage charges and the existing compute cost. Compute is billed separately and unchanged by Genie Paygo. For Genie Spaces specifically, the FAQ says every user gets 150 DBUs of free usage every month, covering Genie, Genie Spaces, and Genie Code. Many users never exceed it. So, that should give you an indication. You can also check this community post.

For the second question... if the goal is simply to surface a Genie Space within another website or application, the API is no longer the only option. Databricks now documents a native iframe-based option here... Embed a Genie Space in an external app. That page explains how a workspace admin must first allow the embedding domain, after which a space author can generate iframe code directly from the Share dialog.

The important limitation is that this does not bypass throughput limits by having many spaces. The iframe path is documented as 20 questions per minute per workspace across all Genie Spaces, not per space.

By contrast, the public docs for the Conversation API say the free tier is best-effort five questions per minute per workspace across all Genie Spaces. So if you were avoiding the API because you have a large number of spaces, that concern is valid, but the same general principle still applies... the limit is enforced at the workspace level rather than separately for each space.

So...

If you only need to embed the existing Genie experience inside another app, use the documented iframe approach. Embed a Genie Space in an external app.
If you need a fully custom UX, custom orchestration, or application-controlled rendering, then the API is still the right path. Use the Genie Spaces API.
If you go down the API route, the docs explicitly recommend retry logic and exponential backoff to handle throughput constraints gracefully.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.

Genie spaces

prasuanu1222 — Mon, 22 Jun 2026 14:09:58 GMT

Hello Team

I am planning to use a genie space for our project as we have seen some changes to costing i would like to understand few things
1) what model does a genie space use and how to get costing estimate for the genie space?
2) we want to embeded this genie space in another application i see we have only api way to do that but we have limited number of api we have very large number of api spaces i dont think that is option how can we find a work around that?