Agent Bricks
Databricks' no-code system for building and auto-optimizing domain-specific AI agents, now generally available with synthetic eval-data generation and multi-agent orchestration.
Overview
Databricks announced general availability of Agent Bricks on October 8, 2026. Instead of generic benchmarks, Agent Bricks auto-generates synthetic, domain-specific evaluation datasets from a customer's own data, then searches across model choice, prompting, and fine-tuning configurations to optimize each task for a chosen cost/quality trade-off. The GA release adds a Multi-Agent Supervisor mode for orchestrating specialized sub-agents (Knowledge Assistant, Information Extraction, and Custom LLM agent types), with MLflow 3 tracing and evaluation dashboards and Unity Catalog governance applied automatically to agent outputs.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Auto-generates synthetic, domain-specific evaluation datasets from a company's own data
- Automatically searches model/prompt/fine-tuning configurations for a chosen cost-vs-quality target
- Multi-Agent Supervisor mode orchestrating Knowledge Assistant, Information Extraction, and Custom LLM sub-agents
- Built-in MLflow 3 tracing and evaluation dashboards
- Unity Catalog governance applied automatically to agent outputs
Use cases
- Enterprise support and knowledge assistants grounded in internal documents
- Structured extraction from unstructured enterprise documents
- Regulated-industry agent deployments needing governance and an audit trail
- Teams without ML expertise building task-specific agents
Pricing
Consumption-based on Databricks compute (DBUs); no separate Agent Bricks license fee.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 8 Oct 2026
- Record updated
- 14 Oct 2026
- Record last checked
- 14 Oct 2026