Agentic AI systems are being adopted by a growing number of organizations. They increase productivity and free people from repetitive tasks. As these systems continue to mature and move into production, companies will gain tremendous value from their ability to work autonomously and make better decisions on our behalf.
However, as the adoption of agent AI grows, new challenges are emerging. One is the lack of skilled talent. Building and managing effective agentic systems requires deep technical expertise, and the demand for experienced AI engineers is rapidly increasing. Another challenge for experts is that it is becoming difficult to keep up with the continuous development of technologies, frameworks and tools in this field.
Kasal was created to address these challenges. It is an agent-driven platform that allows users of various skill levels to design, develop, and deploy effective agentic AI systems through an intuitive visual interface. Non-experts can use Castle’s intuitive UI to build sophisticated agentic AI systems tailored to their specific needs. Experts can use Castle to get started quickly and export their agents to code for deeper customization and development later.
Kasal aims to democratize agentic AI for both experts and non-experts in enterprise environments.
what is castle
Castle is a UI-first framework for designing, running, and observing single and multi-agent workflows. Instead of manually writing complex orchestration code, you can drag and drop agents onto a visual canvas or simply describe what you want through a conversational assistant. Castle will automatically build the workflow for you. You can then connect the tools, run the agents, and observe their behavior in real time. Behind the scenes, Castle uses CrewAIAn open-source Python framework for creating and orchestrating AI agents, but wrapping it in a Databricks friendly application layer that manages authentication, deployment, and monitoring. This means that the flow you design visually can be taken to production with minimal effort. The generated flow can also be exported as code, allowing AI engineers to further refine or extend it as needed.
Why does Castle matter?
Castle brings three core capabilities to the table: a visual workflow designer powered by agents, deep integration with Databricks, and an extensible toolkit that includes MCP servers, Genie, custom APIs, and data connectors.
- visual orchestration Agents, handoffs, and branching make the logic clear and tangible, making it easy to review with non-developers and audit later. Additionally, the agents embedded within Castle understand user intent when defining what type of agent they want to build and propose a customized design that aligns with industry best practices.
- Databricks Native: Castle can be installed as a Databricks app or accessed through MarketAllows users to achieve workspace authentication and governance instead of managing it on their own. Additionally, the flows generated by Castle natively leverage Databricks features, making them enterprise-ready and production-grade from the start. These include MLflow for tracing and tracking, vector search for in-memory, Databricks for serving apps, Lakebase for transaction logging. Incomplete user authentication and more.
- Extensibility: Castle provides first class support mcp serverGenie Spaces, Agent Bricks, as well as custom APIs and data connectors. It can also export flows as notebooks, providing full transparency into its internal logic and allowing AI engineers to extend and refine the solution beyond the initial version with full flexibility.
Castle’s Live Observability provides dual-layer monitoring for multi-agent AI workflows. Through the Castle frontend, business users can view execution timelines that track workflow status, agent interactions, and task progress. Additionally, MLflow tracing integration allows AI engineers to debug model performance, ML calls, and agent behavior. When deployed on Databricks apps, Castle uses Databricks OBO certification for production-ready SQLite or Lakebase persistence for user isolation and transparent agent operations.
starting from castle
A typical user journey starts by prompting Castle with the specifications of the agent you want to create. For example, you might ask: “Create a plan that will create a pitch deck for our sales representatives to sell our various products tailored to customers.” Castle will then use its signals and larger language models to generate a structured scheme, often hierarchical.
In this example, if the plan is in sequential modeThe agents will move one after the other in a prescribed order. However, if the plan is in a hierarchical modeThis would consist of a manager agent and several sub-agents, each responsible for specific tasks: for example, one that retrieves and analyzes customer data, another that retrieves product data, one that combines the two to produce a story for the pitch, and another that prepares a presentation depicting detailed information and the story.
You can then implement a plan to prepare a product presentation for a specific customer. If you want to modify the workflow, such as experimenting with different models or tools, this can be done easily through the Castle user interface.
If you think the plan you create in Kasl could be valuable to others, you can register it in Kasl’s catalog, making it available for reuse and promotion in future sessions. If you want to industrialize the plan outside of Kasal, you can export its code and build a production pipeline around it. You have full flexibility to extend and integrate the plan into your broader solution architecture.
What can you make from castle
We are already seeing users building a wide range of agents and multi-agent AI systems with Castle. Below are some examples:
- Data Analysis Pipeline: Agents that query, analyze, and visualize data
- Content Creation System: Collaboration agent for research, writing and content creation
- Business Process Automation: Intelligent workflows that adapt and make decisions
- Research and Development: Agents that gather, synthesize, and present insights.
Today, there are two easy ways to get started with Castle:
- install from Market Directly in your Databricks workspace (recommended): One-click setup that connects Castle as a Databricks app. This approach lets you take advantage of a managed interface, built-in governance, and automatic updates. View more details Here.
- Deploy from source: Clone Databricks Labs GitHub treasury and run the provided deployment script. This option is ideal if you want to customize or extend Castle to suit your specific needs.
- see Castle Series For new features and announcements.
