Key takeaways
- Agent Bricks is the Databricks agent developer platform: model and tool access, deployment, memory, tracing, evaluation and governance for agents written in code, with any framework.
- Genie Agents (formerly Genie spaces) are the low-code way to build agents over tables and documents, and a custom Agent Bricks agent can call them as tools; Knowledge Assistant and Supervisor Agent are legacy.
- The Agent Bricks CLI scaffolds, runs and deploys an agent to Databricks' agent runtime on Databricks Apps; Agent Bricks is in Beta.
What Agent Bricks is
Databricks describes Agent Bricks as its agent developer platform. It provides the infrastructure a production agent needs, access to models and tools, deployment, memory, tracing, evaluation and governance, on the same platform as your data in Unity Catalog. It is for developers who build custom agents in code, from one agent to many serving production traffic.
| Part | What Databricks provides |
|---|---|
| Models | Frontier and open models through Unity Gateway, with one API, switchable without changing agent code. |
| Tools and data | Databricks-managed, custom and external MCP servers, and Unity Catalog functions over your governed data. |
| Memory | Managed stores for conversation state and long-term memory, backed by Lakebase. |
| Deployment | Agent Runtime: your agent server on Databricks-hosted serverless compute through Databricks Apps, with durable runs that recover after a restart. |
| Quality | MLflow Tracing of every step, with LLM judges, custom scorers and expert feedback. |
| Governance | Access to models, MCP servers and functions controlled, with guardrails, rate limits and usage tracking in Unity Gateway. |
Agent Bricks, Genie Agents and Genie One
- Genie Agents (formerly Genie spaces): a low-code way to build agents that answer questions over your tables and documents.
- Genie One: the no-code experience for business users to work with data and AI.
- Agent Bricks: for developers writing and deploying their own agent code; a custom agent can call Genie Agents and Genie One as tools.
- Knowledge Assistant and Supervisor Agent are legacy offerings, not available to new customers; Databricks points to Genie Agents for new low-code agents and the Agent Bricks CLI for code.
How an agent is built and deployed
Install
The Databricks CLI, Python 3.10 or above and the Agent Bricks CLI (pip install databricks-agentbricks).
Authenticate
Sign in to your workspace with OAuth and save a named profile.
Scaffold
agentbricks init creates a project from a template, such as LangGraph or the OpenAI Agents SDK, with tests and an optional chat interface.
Declare resources
agent.toml lists the tools, memory, sessions and tracing the agent depends on.
Deploy
agentbricks deploy provisions those resources, grants the agent's service principal access to them and deploys it to the agent runtime.
What drives the cost
Agent Bricks is in Beta and needs no workspace setting to start. The cost follows what the agent uses: the compute that runs it, Databricks-hosted serverless compute through Databricks Apps on the number of instances you choose, and the models it calls through Unity Gateway, where usage is tracked. Databricks lists Agent Bricks rates on its pricing page; its Premium tier on Azure corresponds to the Enterprise tier on AWS and Google Cloud.
How to start
Pick one job
A question or task with clear data in Unity Catalog and a way to check the answers.
Try low-code first
If the job is questions over tables and documents, a Genie Agent may be enough.
Build in code
Use the Agent Bricks CLI when the agent needs your own logic, tools or framework.
Evaluate
Trace and score answers with MLflow and your experts before go-live.
Deploy and watch
Deploy to the agent runtime, then review traces, usage and cost every month.
What to check before you start
- Data and law: which cloud region your workspace uses and where models process data, against your data rules and the law that applies to you, such as the UAE PDPL or the GDPR.
- Beta: Agent Bricks is in Beta, so check the features you need with Databricks before you commit.
- Permissions: what the agent's service principal and each tool may access.
- Language: test Arabic and English with your own data and documents.
How Birbal AI helps
Birbal AI builds agents with Databricks Agent Bricks with engineers based in the UAE, alongside Databricks Genie Agents and Unity Catalog governance. Birbal AI is an independent company, not part of Databricks.
Questions and answers
What is Databricks Agent Bricks?
Databricks' agent developer platform for building agents in code and running them in production, with models, tools, memory, tracing, evaluation and governance next to your data.
What is the difference between Agent Bricks and Genie Agents?
Genie Agents are low-code agents that answer questions over your tables and documents. Agent Bricks is for writing and deploying your own agent code, which can call Genie Agents as tools.
Which frameworks does Agent Bricks support?
Any framework or harness, Databricks says. The CLI has templates for LangGraph and the OpenAI Agents SDK, and models come through Unity Gateway.
Is Agent Bricks generally available?
Databricks lists it as Beta, with no workspace setting needed to start; you begin by installing the Agent Bricks CLI.
What happened to Knowledge Assistant?
Databricks calls Knowledge Assistant and Supervisor Agent legacy offerings, no longer available to new customers.
Does Birbal AI build Agent Bricks agents?
Yes, with engineers based in the UAE. Ask for a free estimate and get a scope, timeline and cost range within 48 hours.
Sources
- Agent Bricks FAQ (Databricks documentation)
- Agent Bricks quickstart (Databricks documentation)
- Agent Bricks CLI (Databricks documentation)
- Agent Runtime (Databricks documentation)
- Build AI agents on Databricks (Databricks documentation)
- Develop agents on Databricks (Databricks documentation)
- Agent Bricks pricing (Databricks)
Birbal AI is an independent company based in Abu Dhabi. Facts about third parties come from the sources above; check them for the latest position.