Databricks, a San Francisco-based data and AI software company, said on June 15 that it introduced Omnigent, an open-source meta-harness for a common orchestration layer for AI agents across models, frameworks, and development environments.

Omnigent can supervise Claude Code, Codex, Pi, and custom YAML-defined agents in the same session.

It also lets teams apply policies that can pause risky actions, cap spending, limit tool access, support live collaboration, and run sessions in cloud sandboxes.

Databricks said Omnigent is available in alpha and released under the Apache 2.0 license.

The launch comes as companies push to govern growing numbers of AI agents.

Boomi said 82% of organizations expect to fully integrate AI agents into workflows, raising concerns about monitoring, policy enforcement, and cost management.

Omnigent also enters an existing open-source category that includes LangGraph, which describes itself as a Massachusetts Institute of Technology-licensed orchestration framework for single-agent, multi-agent, and hierarchical workflows with human-in-the-loop controls.

Databricks already markets Agent Bricks as its production AI agent platform with multi-model access and governance.

However, the company’s cited product and documentation pages do not explicitly place Omnigent within that stack.

🔗 Source: Databricks

Recent Databricks developments