What is Omnigent?

Summary

• Omnigent is an open-source meta-harness for AI agents. It gives you a common layer over Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, and agents you write yourself, so you can swap or combine harnesses without rewriting, enforce policies and sandboxing, and collaborate on a live session. It is released under Apache 2.0.
• A meta-harness sits above the harnesses you already use. A harness is the runtime that executes an agent loop. Your sessions, policies, and skills stay with you when the harness or model underneath changes.
• The project is organized around three ideas: Composition, Control, and Collaboration.
• Databricks provides a managed version, which is in Beta. It includes a Databricks-operated server that integrates with your workspace identity provider, model access through the Foundation Model APIs and Unity Gateway, and Databricks Sandboxes. The open-source project's README lists its status as alpha.

What is Omnigent?

Omnigent is an open-source meta-harness for building and running AI agents. A harness is the runtime that executes an agent loop, for example the loop inside Claude Code or Codex. A meta-harness sits one layer above: it wraps the agents you already use and makes them interoperable parts of a larger system. Omnigent is released under the Apache 2.0 license and built by the Databricks AI team, Neon, and the Omnigent contributors. The project's README lists its status as alpha.

What problem does a meta-harness solve?

The Databricks launch blog (June 13, 2026) describes the gap. Each harness has its own interface, context, controls, and way of running, so combining or swapping harnesses is hard, and none of it carries over when you change tools. Omnigent adds a common interface above both terminal coding agents and agent SDKs, based on one observation: however a harness calls its model internally, the user-facing shape is the same, messages and files in, text streams and tool calls out.

What do Composition, Control, and Collaboration mean?

Pillar	What it gives you
Composition	Combine multiple models, harnesses, and techniques without rewriting code, and switch between Claude Code, Codex, Pi, and your own agents with one-line changes.
Control	Stateful, contextual policies track agent actions and enforce guardrails such as cost budgets and permissions at the meta-harness layer, not through prompts.
Collaboration	Share live agent sessions by URL so teammates can review, comment, and steer together in real time. Sessions are reachable from the terminal, web, a native app, mobile, and a REST API.

What does an Omnigent agent look like?

An agent is a short YAML file. This trimmed example from the README sets the harness under executor:

name: my_agent
prompt: You are a helpful data analyst.

executor:
  harness: claude-sdk

Run it with omnigent run path/to/my_agent.yaml. To run the same agent on a different harness, change the harness line. The Databricks docs describe the effect: swap one line to change the harness or model while your tools, prompts, skills, and policies stay the same.

How do policies control what an agent can do?

Policies decide what an agent may do, such as running shell commands, editing files, or spending tokens. Each policy returns one of three decisions: allow, ask for approval, or deny. Admins can set policies server-wide, developers per agent, and users per session, with session rules checked first. Built-in policies include asking before shell and file actions, capping tool calls per session, and enforcing a cost budget.

How does Databricks run Omnigent?

Databricks provides a fully managed version of Omnigent, currently in Beta. According to the Databricks docs it includes:

• A Databricks-operated Omnigent server that integrates with your workspace's identity provider.
• Model access through the Foundation Model APIs and Unity Gateway.
• Databricks Sandboxes for secure, collaborative agent coding and knowledge work. Databricks Sandbox is in Beta and available in select regions on AWS.

On Databricks, Omnigent uses your workspace identity, with no separate Omnigent login or password.

	Open-source Omnigent	Omnigent on Databricks
Stage	Alpha (README status)	Beta
Server	You deploy and run it, for example with Docker Compose or on Databricks Apps	Operated by Databricks
Sign-in	Built-in accounts or OIDC	Your Databricks workspace identity
Model access	API key, subscription, gateway, or a Databricks workspace	Foundation Model APIs through Unity Gateway; no bring-your-own API keys on a Databricks Sandbox host
Custom policies	Python policies registered on the server	Custom policies written in CEL; custom Python policies are not supported

What do I need to use the managed version?

A workspace admin must enable the Omnigent preview, and the workspace must be in a region that supports Unity Gateway. Sessions on a Databricks Sandbox also need a region that supports Databricks Sandbox and the Sandbox preview enabled. Do not use Databricks Sandbox in a workspace with serverless egress control (SEG) enabled.

FAQs

Do I need Databricks to use Omnigent?

No. The open-source project can be self-hosted, and its README lists an API key, a subscription, or a compatible gateway as model credentials. A Databricks workspace is one option among several.

Does Omnigent replace Claude Code or Codex?

No. It sits above them. Omnigent runs those agents as harnesses and adds sharing, policies, and sandboxing on top, as the launch blog describes.

Can I bring my own model API keys on Databricks?

Not on a Databricks Sandbox host, where model access routes through Unity Gateway. On a machine you register yourself, the quickstart lets you route a harness through Unity Gateway or provide your own API key, subscription, or gateway.

Is Omnigent generally available?

No. The open-source project is alpha and the Databricks managed version is Beta. For current availability, see the Databricks Omnigent docs and the open-source documentation.