# Agent Bricks. Production-ready AI agents, on your data.

![Agent Bricks interface showing multi-model routing, evaluations, and secure data access.](https://developers.databricks.com/img/products/hero/agent-bricks.png)

The agent platform with multi-model routing, built-in evals, and secure data access.

- [Build apps with Agent Bricks](https://developers.databricks.com/templates/ai-chat-app)
- [Read the docs](https://developers.databricks.com/docs/agents/overview)

## Build the agent, not the agent stack.

One environment for building, testing, and running agents — without stitching together separate systems.

### Idea to agent

Start from a prompt or template and iterate immediately, without setting up infrastructure first.

### Less overhead

Focus on agent logic instead of wiring services, integrations, or maintaining supporting systems.

### Production-ready

Go from prototype to production with the same setup, without rebuilding as you scale.

## What you get with Agent Bricks, out of the box

### Multi-model routing: Use the right model for every task.

Switch between leading models like GPT, Claude, Llama, and more.

Send requests through a single API and route across models by cost, performance, or availability — without building your own orchestration layer.

- Access models from multiple providers in one place
- Route requests by cost, performance, or availability
- Built-in fallback and usage controls

### Built-in evals: Control and improve output quality.

Measure how your agent performs and make targeted improvements.

Run your agent against real scenarios, evaluate responses, and improve based on clear feedback — all in one place.

- Generate eval datasets from real use cases
- Score outputs for quality, relevance, and correctness
- Compare results across prompts, models, and iterations

### Secure data access: Use your data, with the right permissions.

Access real data securely — without copying or bypassing controls.

Connect your agent to governed data and run queries with proper access controls, so it only sees what it's allowed to see.

- Query data with user-level permissions
- No data duplication or reverse ETL
- Enforce access controls and governance by default

## Built for real-world agents.

Common ways teams use Agent Bricks to build and ship AI-powered applications.

### AI Copilots on Your Data

Assist users by querying real data, generating responses, and taking actions with the right permissions.

### Customer Support

Handle requests with agents that retrieve context, generate replies, and improve over time.

### Internal Automation

Automate workflows by connecting systems and executing tasks beyond simple text generation.

### Decision-Making Systems

Build systems that analyze inputs and take the next best action in real time.

### Content Pipelines

Generate and validate content at scale with built-in evaluation to ensure consistent output quality.

### RAG Systems

Ground responses in your data while enforcing access controls and permissions.

### Experimentation

Compare prompts and models to optimize output quality, cost, and performance.

### Background Tasks

Run agents asynchronously to process tasks, analyze data, and act without user interaction.

## Agent Bricks powers real agents in production.

See how teams ship AI applications on governed data.

> "With Agent Bricks, our teams were able to parse through more than 400,000 clinical trial documents and extract structured data points — without writing a single line of code. In just under 60 minutes, we had a working agent that can transform complex unstructured data usable for Analytics."
>
> Joseph Roemer, Head of Data & AI, Commercial IT, AstraZeneca

> "Agent Bricks enabled us to double our medical accuracy over standard commercial LLMs, while meeting Flo Health's high internal standards for clinical accuracy, safety, privacy, and security."
>
> Roman Bugaev, CTO, Flo Health

> "With Agent Bricks, we can quickly productionize domain-specific AI agents for tasks like extracting insights from customer support calls — something that used to take weeks of manual review."
>
> Chris Nishnick, Director of AI, Lippert
