databricks Databricks Blog ·

Databricks and Omnigent streamline AI agent development with Nimble web search

blogaidatabricksgaengineer
feature

Databricks users can now leverage Omnigent, a new layer for defining AI agents once, to simplify development and management. Omnigent integrates with Databricks Foundation Model APIs for unified cost tracking and governance, addressing the inefficiencies of building agents across multiple harnesses. By utilizing Nimble for specialized web search, agents achieve higher accuracy and reduce search costs. This solution helps engineers focus on agent enrichment rather than plumbing, improving the consistency and reliability of data-driven insights.

  • Unified Agent Definition and Management with Omnigent
  • Problem: Inconsistent Agent Development and Web Search
  • Specialized Web Search via Nimble Integration
Features (1)
  • Unified Agent Definition and Management with Omnigent

    Omnigent allows engineers to define an agent's model, tools, policies, and limits once, then run it across various harnesses. This integration routes model calls through Databricks Foundation Model APIs, providing unified cost visibility, audit trails, and governance across all agent activities.

Enhancements (1)
  • Specialized Web Search via Nimble Integration

    Omnigent integrates with Nimble's specialized web search API to provide consistent, high-quality real-time web data for agents. Nimble's web search agents improve LLM benchmark accuracy from 46% to 71% and cut search costs in half by adapting to specific use cases and efficiently retrieving relevant data.

Notes (1)
  • Problem: Inconsistent Agent Development and Web Search

    Engineers often rebuild AI agent logic multiple times across different execution environments, leading to wasted effort and inconsistent results. Bundled web search tools in these harnesses also provide incomplete data, lack shared cost tracking, governance, and audit trails.

Read the original announcement →

https://www.databricks.com/blog/web-search-your-agent-inherited-isnt-good-enough

Related releases