Databricks Unveils Omnigent to Streamline AI Agent Search

Developers building AI agents often face the tedious task of recreating the same logic across multiple environments, such as Claude Code, Codex, and raw APIs. This fragmentation wastes engineering hours on repetitive plumbing and results in inconsistent web search outputs. To solve this, Databricks has introduced Omnigent, a new layer that allows developers to define an agent's model, tools, and policies a single time and run it seamlessly across any of these harnesses.

By consolidating these builds into a single definition, Omnigent routes all model calls through the Databricks Foundation Model APIs. This integration provides a unified system for tracking costs, auditing performance, and enforcing governance policies, which previously had to be managed separately for each runtime. To deploy this setup, users must enable the Omnigent Beta in their workspace and ensure their region supports the Unity AI Gateway.

A critical component of this new architecture is the integration of Nimble as the dedicated web search provider. Standard search tools bundled with LLM harnesses often miss granular details, but Nimble specializes in deep domain retrieval, bypassing JavaScript and pagination to find elusive data. In benchmark testing, replacing default search with Nimble's Search API raised LLM accuracy from 46 percent to 71 percent, while simultaneously cutting web search costs in half compared to Claude's native search.

For practitioners managing large-scale tasks, such as monitoring hundreds of thousands of customer accounts for market signals, this combination bridges the gap between internal and external data. Internal records stored in Databricks Delta tables governed by Unity Catalog can now be seamlessly combined with real-time web insights. By making a simple one-line configuration change to designate Nimble as the default search tool, developers can ensure their agents deliver consistent, auditable, and highly accurate results.