Omnigent is a layer that lets engineers define an agent once — model, tools, policies, limits — and run it across any harness (Claude Code, Codex, raw API), with Nimble filling the web search slot. Building the same agent multiple times across different harnesses wastes engineering time on plumbing, and each harness's bundled web search returns inconsistent, incomplete results with no shared cost tracking, governance, or audit trail. One agent definition replaces three rebuilds, model calls route through Databricks Foundation Model APIs for unified cost and governance, and Nimble's web search raised benchmark accuracy from 46% to 71% while cutting search costs in half. An agent that needs the outside world An engineer at a software company is building an agent to keep the company's view of the market up to date. It monitors a few hundred thousand prospects and customer accounts for signals that an account is open to engagement: a new funding round, a leadership change, a product launch, or a hiring surge that indicates budget. The account records already live in Databricks, in Delta tables governed by Unity Catalog and joined to the company's own usage and pipeline data. …