From the source
Databricks announced Omnigent, a layer that allows engineers to define an agent once and run it across multiple harnesses such as Claude Code, Codex, and raw API, with Nimble providing web search capabilities.
This approach replaces rebuilding the same agent for each harness, routes model calls through Databricks Foundation Model APIs for unified cost and governance, and claims to raise benchmark accuracy from 46% to 71% while cutting search costs in half.
