New Agent Bricks documentation and compute stack on Databricks
Databricks updated Agent Bricks documentation focusing on code-first agent development with new server and runtime options.
O que mudou
Use cases
Imagine you need to deploy a custom AI agent into production with support for durable executions and crash recovery. With the updated Agent Bricks documentation and architecture, you can:
- Build and deploy agents using code.
- Ensure crash recovery and synchronous, streaming, or background runs using
DurableAgentServer. - Run agents at scale using Agent Runtime on Databricks Apps.
- Query deployed agents through dedicated APIs.
How to use
Use the Agent Bricks CLI to scaffold and structure your agent code locally before deploying.
# Scaffold and create your agent project using the CLI
databricks agents create my_agent
What happens in this code:
- The command-line tool initializes the required file structure for agent development.
- The generated baseline code connects to managed Databricks services for local iteration.
Limitations and requirements
Legacy offerings such as MLflow AgentServer, LongRunningAgentServer, and agents deployed to Model Serving have been moved to a legacy section and are no longer recommended for new agent deployments.
- mosaic ai
- mlflow
- databricks apps
- unity catalog
Agent Bricks documentation for code-first agent development
Resumo e explicação gerados por IA a partir da release note oficial. Em caso de dúvida, a documentação oficial da plataforma prevalece.
Relacionadas
- DatabricksGA
Repositórios de pacotes em nível de workspace para Databricks Apps
Administradores de workspace podem configurar repositórios padrão de pacotes Python e npm para Databricks Apps.
databricks apps · workspace settings - DatabricksGA
Redimensione partições de stateful streaming sem perder o checkpoint no Databricks
Agora é possível alterar o número de partições de shuffle em queries Structured Streaming e pipelines Lakeflow sem descartar o estado do checkpoint.
structured streaming · lakeflow - DatabricksAtualização
Google Gemini 4 Argon disponível no Unity Gateway do Databricks
O modelo multimodal Google Gemini 4 Argon já pode ser acessado via Foundation Model APIs e Unity Gateway.
unity catalog · foundation model apis