Databricks Releases Omnigent, an Open-Source Meta-Harness for AI Agents

Databricks has released Omnigent, an open-source 'meta-harness' for AI agents, under the Apache 2.0 license. The Databricks AI team built it in collaboration with Neon. A harness is a wrapper around a model that turns it into an agent, and Omnigent sits above multiple harnesses, treating each as an interchangeable part of a larger system.

Many engineers currently juggle multiple agents at once, copying text between coding agents, search tools, and communication platforms. Each harness only understands its own sessions, but Omnigent adds a shared layer for composition, control, and collaboration. It provides a common interface above command-line agents and agent SDKs, wrapping terminal coding agents like Claude Code, Codex, and Pi, as well as SDKs like OpenAI Agents and the Claude Agents SDK.

The design of Omnigent is based on the observation that, regardless of how a harness calls its model internally, the user-facing interface remains the same. Messages and files go in, and text streams and tool calls come out. Omnigent standardizes this interface, making harnesses swappable.

Users supply the models and infrastructure, and Omnigent runs the agents on top, coordinating multiple agents as interchangeable workers under one orchestrator. The architecture of Omnigent consists of two parts: a runner that wraps any agent in a sandboxed session with a uniform API, and a server that provides policies and sharing. The server exposes every session over the terminal, app, and web APIs.

A single command starts a session in the terminal and launches a local web UI at localhost:6767, which appears in sync across the browser and phone. The Databricks team frames Omnigent around three key capabilities. An OS sandbox, called Omnibox, underpins this, locking down OS access and transforming network requests.

Two example agents ship with the repository, demonstrating practical patterns such as a frontier advisor model guiding a cheaper open-source worker. The Marktechpost team has created an interactive demo that lets users experience Omnigent's meta-harness workflow firsthand. The demo showcases an orchestrator managing multiple harnesses in a governed session.

Why this matters: The release of Omnigent has significant implications for the AI and development communities. By providing a standardized interface for AI agents, Databricks is enabling developers to build more complex and collaborative AI workflows. This could lead to increased adoption of AI in industries such as software development, customer service, and research.