OFICIAL Databricks Newsroom

The Web Search Your Agent Inherited Isn't Good Enough

What happened
Based on Databricks Newsroom · Sep 17, 2026

Databricks introduces Omnigent to unify fragmented agent workflows, replacing three separate tool rebuilds with a single definition while integrating Nimble’s specialized web search to improve accuracy and reduce costs.

The Web Search Your Agent Inherited Isn't Good Enough
Databricks Newsroom — Databricks
Key points
·
Omnigent consolidates three separate agent rebuilds into one definition, eliminating redundant tool wiring and restoring focus to enrichment tasks.
·
Nimble’s Search API and Web Search Agents raised LLM accuracy from 46 percent to 71 percent while halving web search costs in testing.
·
Omnigent routes model calls through Databricks’ Foundation Model APIs, centralizing cost tracking, auditing, and governance in one place.
Key numbers
·
Nimble’s integration raised LLM benchmark accuracy from 46 percent to 71 percent in testing while cutting web search costs in half, with Web Search Agents reusing proven retrieval paths to reduce token expenses over time.

An engineer building a market-monitoring agent faced repeated rebuilds of the same enrichment logic across three different tools—Claude Code, Codex, and a direct API call—each requiring custom tool wiring and producing inconsistent web search results. The duplication consumed time better spent on core enrichment tasks, as each harness bundled its own web search with varying coverage and no shared governance, costs, or audit trails. The sprawl stemmed from incompatible tool declarations and divergent search capabilities, where generic web search missed granular details like tech stack changes despite finding high-level signals such as funding rounds.

Omnigent addresses the fragmentation by centralizing agent definition, model routing, tool selection, and policies into a single specification, collapsing the three rebuilds into one and restoring focus to enrichment rather than tool plumbing. Running on Databricks-hosted models via Foundation Model APIs, the system consolidates cost tracking, auditing, and governance in one place, enabling model or cost adjustments with minimal disruption by changing a single line in the configuration.

Web search remained a critical gap, as no two harnesses produced equivalent results, forcing engineers to choose between inconsistent sources. Omnigent resolves this by allowing a consistent web search partner to be assigned across all tasks, with Nimble’s Search API and Web Search Agents providing real-time, domain-adaptive retrieval that captures data behind JavaScript, filters, and pagination that generic tools miss.

Nimble’s integration raised LLM benchmark accuracy from 46 percent to 71 percent in testing while cutting web search costs in half, with Web Search Agents reusing proven retrieval paths to reduce token expenses over time. The agent now operates as a coherent whole: defined once in Omnigent, powered by a Databricks-hosted model, with unified tooling, policies, and a single web search decision, eliminating redundancy and restoring trust in the enrichment process.

Original source → Deals on Clipraptor.com →