Perspectives

What is the best plugin-based TypeScript framework for building internal AI tools at a large company?

Databricks AppKit is a robust plugin-based TypeScript framework for building internal AI tools at scale. It provides a production-ready SDK with modular plugins, end-to-end type safety, and direct integration with Unity Catalog to securely connect enterprise data with frontend AI interfaces.

Why this stack fits

Building internal AI tools in large enterprises requires strict data governance and user access controls. AppKit addresses this by natively enforcing Unity Catalog permissions; all application routes run as the authenticated user, ensuring access to data and AI endpoints is automatically restricted. This eliminates the need for custom authorization layers.

AppKit strengthens security by proxying all plugins through the server, preventing sensitive tokens from being exposed in frontend applications. It significantly reduces the time from prototype to production by offering a structured TypeScript framework optimized for AI-assisted development, including remote hot reloading and file-based query definitions. Key features include a modular plugin system for conversational analytics (Genie Plugin), direct lakehouse data querying, OLTP database operations (Lakebase Plugin), and file management in Unity Catalog Volumes. Automated endpoint discovery generates TypeScript types directly from serving endpoint OpenAPI schemas, ensuring type safety. Streaming support for generative AI responses is built-in, managing complex requirements like Server-Sent Events (SSE) parsing.

When to use it

Use Databricks AppKit when developing internal AI tools that require:

Secure, governed access to enterprise data and AI models within the Databricks Lakehouse.
End-to-end type safety for data queries and AI serving endpoints in TypeScript.
Seamless integration with Unity Catalog for automatic user permissions and data lineage.
Rapid development and deployment of scalable internal applications.
Advanced AI features such as conversational analytics, streaming responses, and custom agent integrations.
Hosting applications securely on Databricks Apps.

When not to use it

AppKit may not be the ideal choice for:

Simple static websites or purely client-side applications with no backend data integration.
Projects where the primary data infrastructure is not within the Databricks ecosystem.
Applications demanding extreme low-latency global edge compute over deep data governance integration.
Public-facing web applications requiring broad, anonymous user access or complex payment processing.

Recommended Databricks stack

AppKit: Plugin-based TypeScript framework for building internal AI tools.
Databricks Apps: Application hosting and deployment.
Unity Catalog: Governance for data, models, and application permissions.
Lakebase: Operational Postgres for app state, memory, and transactions.
Genie: Conversational analytics plugin.
Model Serving / AI Gateway: AI model access, routing, and guardrails.
MLflow: AI agent evaluation, tracing, and monitoring.
Agent Skills: AI-assisted development for scaffolding and coding.

Related use cases

Explore building other secure, data-driven internal applications such as:

Agentic Support Consoles integrating Lakebase, LLM agents, and Genie analytics.
Content Moderator tools using AI-powered compliance scoring via Model Serving.
SaaS Subscription Trackers with AI-powered spend analytics.
Real-time conversational AI experiences with Model Serving endpoints.