Databricks has introduced Omnigent, an open-source meta-harness that lets organisations combine, govern and collaborate on AI agents spanning different models, frameworks and development environments. Released under the Apache 2.0 licence and available now in alpha, Omnigent positions itself as an orchestration layer sitting above the patchwork of agent systems enterprises increasingly run side by side.

The launch targets a growing pain point in enterprise AI. As teams deploy agents to automate workflows, generate code and support decision-making, they often end up managing multiple agent systems that operate in isolation. Omnigent connects these through a common interface, applying governance, security and collaboration controls consistently across environments rather than per tool.

Bringing order to multi-agent enterprise AI

Omnigent lets developers integrate agents built with tools such as Claude Code, Codex and Pi, alongside custom frameworks, into a single system. The harness is organised around three core capabilities. Composition allows teams to combine agents, models and frameworks through one interface and switch between them with minimal code changes. Control applies stateful governance, security and cost-management policies that track agent actions and enforce guardrails beyond simple prompt-based instructions. Collaboration lets teams share live agent sessions, review outputs together and steer workflows in real time.

Additional features include cloud execution environments, operating system sandboxing, contextual security policies and cost budgeting controls, along with support for building bespoke multi-agent systems across frameworks.

Built on Databricks’ own agent deployments

The company said Omnigent draws on its experience running AI agents across its global engineering organisation and building agent-powered products such as Genie. That work surfaced a recurring pattern — the most advanced AI workflows increasingly involve multiple models, multiple frameworks and multiple users working together, creating demand for a higher-level system to coordinate and govern those interactions.

Databricks describes itself as the data and AI company, with more than 20,000 organisations worldwide using its platform, including 70 per cent of the Fortune 500. The release adds an agent-orchestration layer to a portfolio that already spans Lakebase, Genie, Agent Bricks, Lakeflow, Lakehouse and Unity Catalog.

For enterprises in Singapore and across the region weighing how to scale agentic AI without losing oversight, an open-source governance layer that is framework-agnostic could lower the barrier to consolidating fragmented agent deployments under consistent controls.