Inside Omnigent: Databricks' New Open-Source Meta-Harness Composing, Controlling, and Sharing AI Agents

Databricks has released Omnigent, an open-source project it calls a "meta-harness" designed to solve a critical pain point for developers: the fragmentation of AI coding agents. Instead of replacing existing tools like Claude Code, OpenAI Codex, or Pi, Omnigent installs a layer above them. This creates a single unified interface where work, context, and controls are preserved when a developer needs to switch from one agent to another .

What is a Meta-Harness and Why Does It Matter?

In today's AI landscape, a "harness" is the software wrapper that turns a large language model into a functional agent that can interact with tools, write code, and execute tasks. The problem, as described by Databricks and early users, is that each harness is its own silo. Claude Code remembers its own sessions and rules, but that context vanishes the moment a developer needs to use Codex or Pi for a specific sub-task .

Omnigent, released as an Apache 2.0 open-source project on June 13, 2026 , is not another coding agent. It is a "meta-harness" that sits above the agents you already use and provides them with a shared runtime layer for composition, control, and collaboration . You bring your own models, your own API keys, and your own infrastructure; Omnigent keeps your sessions and policies constant regardless of which agent is running underneath .

The Three Core Pillars of Omnigent

Omnigent's mission to end agent fragmentation is built on three foundational capabilities :

1. Composition
Omnigent allows developers to combine multiple agent harnesses and large language models without rewriting their entire workflow. This includes creating multi-agent teams where different harnesses handle different parts of a task, or even swapping one model for another mid-session and mid-loop with simple, one-line configuration changes .

2. Control
Without a shared layer, governance policies like spending limits or access rules must be re-implemented for every tool. Omnigent introduces stateful, data-centric policies and cost budgets that are enforced at the meta-harness level, ensuring that rules follow the work, not the specific tool being used .

3. Collaboration
Perhaps the most immediate benefit is the ability to share live agent sessions. A developer can start a complex task on a desktop terminal, pick up the exact same session on a phone, and share it with a teammate via a URL for real-time collaboration .

Getting Started: Licensing, Installation, and Repository

Omnigent is fully open-source and released under the permissive Apache 2.0 license . It is currently in an alpha state, meaning it is available for developers to use and contribute to, but may not yet be feature-complete or fully stable .

• License: Apache 2.0 .
• GitHub Repository: github.com/omnigent-ai/omnigent .
• Quickstart Installation: A quickstart guide is available at omnigent.ai/quickstart/install .
• Official Documentation: Full docs are hosted at omnigent.ai .

The project was introduced by Matei Zaharia, the original creator of Apache Spark, and was developed in collaboration with Neon .

Release Timing and Data + AI Summit 2026

Omnigent was open-sourced on June 13, 2026 , just two days before Databricks’ flagship Data + AI Summit 2026, which runs June 15–18 at the Moscone Center in San Francisco . This release cadence immediately makes Omnigent a key topic of discussion at the world's largest gathering of data and AI professionals.

Omnigent vs. Agent Bricks: Two Distinct Tools for Different Needs

A common point of confusion is how Omnigent relates to Databricks' enterprise platform, Agent Bricks. The short answer: they are complementary but target entirely different users and problems. Omnigent is a lightweight, decentralized developer tool for coordinating multiple third-party agents. Agent Bricks is a centralized, governed enterprise platform for building and deploying agents directly on business data.

In essence, Omnigent solves the problem of orchestrating the external AI tools a developer already uses, while Agent Bricks solves the problem of building and governing AI agents for business applications in a secure, governed manner .

Dimension	Omnigent	Agent Bricks
Type	Open-source meta-harness.	Databricks enterprise agent platform.
Target User	Developers orchestrating multiple external agent tools.	Enterprise teams building governed agents on business data.
Scope	Provides a single UI layer over Claude Code, Codex, Pi, and custom YAML-defined agents.	Unifies model access, execution, governance, and context in one system for business-data-aware agents.
Licensing	Apache 2.0.	Not specified in provided sources; follows Databricks' usage-based pricing with no upfront license fee.
Key Focus	Composition, control, and collaboration across heterogeneous tools.	Enterprise governance, identity, model access, and automated optimization on business data.
Where it Runs	Your laptop, terminal, web, desktop, and phone.	The managed Databricks Data Intelligence Platform.

Agent Bricks is an end-to-end solution designed to move AI agents from pilots to production. It leverages the Mosaic AI research engine to automate agent creation, evaluation, and optimization, all while maintaining enterprise-grade governance through Unity Catalog . Omnigent, meanwhile, is the portable, bring-your-own-keys layer that ensures a developer’s context and policies are no longer held hostage by whichever coding assistant is trendy this month.