Loading Databricks updates...
← All video guides
Generative AI & Agents

Kasal: Databricks Labs Low-Code Agent Framework with Genie Integration

Kasal is a Databricks Labs field engineering project enabling non-technical users to build multi-step AI agents in natural language.

In short

Kasal is a Databricks Labs field engineering project enabling non-technical users to build multi-step AI agents in natural language. Deploy from Databricks Marketplace; runs on Databricks Apps with built-in authentication and authorization. Distinct from managed Databricks Agents framework — Kasal prioritizes conversational ease over code flexibility.

Source: official Databricks documentation · Databricks announcement

Key takeaways

  • Kasal is Databricks Labs field engineering low-code framework for building multi-step AI agents.
  • Non-technical users create agents and tasks via natural language prompts; no coding required.
  • Deploys on Databricks Apps with built-in authentication, authorization, and data access respecting UC permissions.
  • Integrates Genie rooms, external APIs, search engines; orchestrates multi-task workflows and generates artifacts.
  • MLflow traces execution; non-developers can iterate without code and save agent plans for reuse.

Chapters

  • 0:00 — What is Kasal? Databricks Labs Low-Code Agent Framework
  • 0:37 — Why Kasal: Enable Non-Technical Users to Build Agents
  • 1:15 — Kasal vs. Databricks Genie: Different Use Cases
  • 2:10 — Kasal Architecture: Databricks Labs Project Deployed on Apps
  • 2:30 — Installation: One-Click Deploy from Databricks Marketplace
  • 3:44 — Accessing Kasal: Launched via Databricks Apps
  • 4:46 — Creating Agents: Natural Language Prompts
  • 5:18 — Building Multi-Step Plans: Tasks and Workflows
  • 6:19 — Assigning Tools to Plans: Genie, APIs, External Sources
  • 6:51 — Connecting Genie Rooms for Data Access
  • 7:22 — Workspaces: Team Organization and Collaboration
  • 7:53 — Tracing Execution: MLflow Visualization of Agent Workflows
  • 8:23 — Iteration Without Code: Modify and Re-run
  • 9:00 — Dashboard Outputs and Shared Artifacts
  • 9:14 — Conclusion: Next Steps for Low-Code Agent Development

Frequently asked questions

What is Kasal?

Kasal is a low-code agent framework from Databricks Labs that enables non-technical users to build multi-step AI agents using natural language. Create agents and tasks by typing prompts; Kasal generates the execution plan.

How does Kasal differ from Databricks Genie?

Genie answers single questions from data. Kasal builds agents that perform sequences of steps—gather data, analyze, create dashboards, send emails. Kasal prioritizes conversational ease over code flexibility.

Can I connect Kasal to my data?

Yes. Kasal integrates Genie rooms (for SQL/analytics), external APIs, and search engines. Data access respects your Unity Catalog permissions—agents only access data you already have permission for.

How does Kasal track and iterate on agent runs?

Execution traces appear in MLflow for visibility. Non-developers can modify prompts and re-run agents without code. Save successful agent plans to a catalog for team reuse.

Full transcript

Auto-generated from the video and lightly edited for readability. Timestamps link to the moment in the video.

Show full transcript (25 segments)

0:04Hey, today I have the pleasure to host Neto. He's a data AI strategist at datab bricks. And if I'm not wrong, he's also the person behind an awesome tool called kasal, which literally means lazy in Arabic. Ne, can you tell us more about you and how you're going to make us lazy?

0:26>> Thank you. Pleasure. Mine, you indeed. Casal is um a word in Arabic that also meant lazy and kasal is helping us to be to be able to build agents in a low code no code approach. So imagine yourself as um a persona that is non-technical in terms of uh you you

0:50don't know how to write code and you want to be able to build agents in a short period of time to be able to provide value. Imagine a persona that wants to do a research on a particular topic and wants to be able to create the agents that would do the research. Now that data could be data that is coming from our genie solution, it could be a

1:12data coming from our agent brick solution or it could be some external data be it APIs that are available or be it information that are from the internet through search engines. Then such a persona would be able to in a no code low code approach prompt that solution and would be able to create several sequence of steps that would be gathering that data resonating that data

1:35and then producing different artifacts be it dashboards be it sending emails be it connecting and doing different things before doing the action. So just to summarize casal is a low code no code solution to build agents. So you're literally saying that in addition to Databricks one where we can democratize access to AIBI dashboard genie and apps

1:57we are also having another solution called casal that will help us build agents quickly. >> Correct. So so um casal is a is a datab bricks labs porn project. So it is a part of the field engineering team. We were building the solution that allows us to deploy Castle in datab bricks apps and within the datab bricks apps we are able to have the authentication

2:20authorization where we're able to connect to the different resources in datab bricks without having to basically run and and uh and do the code. So the difference between datab bricks one where you're able to access dashboards be able to text to SQL in order to gather that different data. Think of it as you're building agents where you're

2:43giving them a goal. They have to go and then fetch different things. Identify for instance the most effective campaign. After they gather those campaign, they want to be able to communicate those results to the brand manager and then afterwards gather these different documents and be able to consolidate them, extract some information from them and then do a next action. So whenever you you do have

3:07several steps that have to happen that integrates with your overall ecosystem that that is really a good use case where you could leverage cas >> and how can I deploy Cassal in my workspace. >> So actually let me share my screen and I'll I'll walk you through how you'd be able to deploy Casal. So there in at at Casal you could go to the datab bricks marketplace. in the marketplace you would search for Casal and once you find

3:30Casal with a point and click you'll be able to deploy Casal in your environment. So that installation mainly will allow you to within just that simple um u point and click you will have the castle within the environment. But since this is installed within the datab bricks apps whenever you want to access uh the castle itself you will

3:54be able to see it available within the apps. So if you go to the compute then to the apps you go filter your own apps you would be able then to see that the castle market that is the one where just we just triggered. So I have another Cassell deployment here where I'd be able to click on the link and notice here I didn't really uh require to do to be redirected to any login

4:17page. This is because I am already logged into my environment. Now that means that while you're in Castle everything that you are able to do on the uh data that is available within datab bricks environment Casal will be able to do that. Everything that you cannot the tables that you don't have access to then your agent won't be able to do that. That means also in a multi-tenency there might be some other

4:39users that have access to the data with Casal. They would be able to have access but you won't be able to do that. How you would be able to create agents. So in this assistant on the right side you could see that you could say something like create a marketing uh strategist agent. So that one single prompt is creating one agent for you. Now think of it as yes this might be your digital FTE

5:03that you want to give it different task that it has to u do it could be going and gathering data from external sources or from Genie from our texts SQL solution. So for instance, if I would say here create a task that will identify the most effective campaign from the past 3 months, that one

5:25particular prompt is going to create that task for me. And think of it as you would be able to create one or many agents, one or many tasks within that agent. And the moment you have a plan that you believe that this is what you would like to have, then you'll be able to trigger that run. So this is what this example that I just showed you was creating one marketing strategist that does the campaign effectiveness. But if

5:49I could actually start here in a new canvas, I would even give one prompt that says create a plan that will gather the most effective marketing campaign and create a presentation from it. So for the sake of time here, I'll say maximum two agents and two tasks. So this is rather than creating one agent

6:12and one uh task at a time, you you just create a whole an entire plan where [clears throat] it will be multiple agents, multiple tasks that will uh go to your marketing tables, gather all the most effective campaigns and then in the end create a presentation for you. You could see see you want to create a dashboard, you want to create a presentation or you want to send an

6:34email that would be following that same pattern. So here what I've done is I was able to create that plan. Now I'm able to assign the tools to this plan. The tools here could be some something like the external source like perplexity or being able to call genie. Now when I call genie I would be able to select my genie space. So here I would be able to

6:56specify the genie room that I want. Once I specify I could up before starting the call go back to genie validate that this is the right room. those were the questions that um that room is able to to answer and then in the end once you validate that you say yes this is the room you save it and then you trigger this run so while you trigger

7:18this run there's this concept of workspaces where you could have a workspace for marketing workspace for engineering for go to market so here for instance I have a a workspace that is for marketing that I am a member of and the moment I changed workspaces I could see all the queries that have been done by either myself or the people part of the workspace. So if I look here, I have

7:40a marketing campaign analyst dashboard that got created. If I look at the trace, I could see that this was a brand manager that was mainly trying to identify the cost and number of leads generated and that was the query going towards Genie and Genie replied back with that particular answer. Now that answer went to an input to another agent

8:02that was visualizing that content in in a nice dashboard. So if I want to look at the result, I'll be able to open the dashboard and then I see that dashboard that is created from that genie query that came back from the first agent. So this is something that I might be happy with. I might want to um say no, I want to be able to iterate on it. As a as a non uh developer persona, I would be

8:26able to easily iterate through on it by uh specifying well this is not enough. I want to be able to change the content out of it and then rerun it easily. And the moment I'm happy with it, I go ahead and I could save it and I say this is the marketing dashboard. And the moment you save it, other members would be able to go on the catalog and then pull on pull in that plan that you've been

8:49you've been creating for that. >> That's that's amazing. And I think I think thanks to you, I'll be adding Asian developer in my LinkedIn profile. That's a short summary. This video is part of a series that will go and deep dive in every single component of Castle to make you an expert or an agent

9:11machine. So stay tuned. Looking forward to

Official Databricks resources

Related

Databricks Answers

Looking for a quick, source-backed answer? Browse Databricks Answers on GenAI, agents & Vector Search — direct, documentation-validated answers to common Databricks questions, curated by Youssef Mrini, Senior Developer Advocate at Databricks.