Building a user-facing interface for an agent, such as an app, Slack, or Teams

For a Generative AI agent to be usable, it needs a front-end that end users can interact with, and Databricks supports multiple deployment paths for exposing an agent beyond notebooks. The choice of interface depends on where the users already work (a custom web UI, a chat platform like Slack/Teams, or an embedded Databricks-native experience) and how the backend agent is exposed (typically via Model Serving endpoints). Regardless of the interface chosen, the underlying agent logic and serving endpoint remain the same, so only the presentation layer changes.

1 · Learn the must-know

• 1Databricks Apps allow you to build and host lightweight, interactive web UIs (e.g., using frameworks like Streamlit, Gradio, or Dash) directly within the Databricks workspace, calling a Model Serving endpoint on the backend.
• 2Agents deployed to Model Serving expose a REST API endpoint, which is the common integration point for any external interface, including Slack bots, Microsoft Teams apps, or custom web/mobile clients.
• 3Integrating with Slack or Teams typically requires building a middleware/bot layer (e.g., using each platform's SDK/webhook framework) that receives user messages, calls the Model Serving endpoint, and returns the formatted response.
• 4A key gotcha is authentication: interfaces calling a Model Serving endpoint need a valid Databricks personal access token or service principal token, and this must be managed securely (e.g., via secrets) rather than hardcoded in the app.
• 5Choice of interface should match the usage scenario: internal experimentation or demos often use Databricks Apps or notebooks, while production business workflows favor embedding into existing collaboration tools like Slack/Teams where users already operate.
• 6Streaming responses (token-by-token) from the serving endpoint can improve perceived responsiveness in chat-style interfaces, but the interface must be built to handle streaming output if the endpoint supports it.

2 · Check your understanding