Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex

---

What is Omnigent — Databricks’ Meta-Harness Abstraction Layer for Agents

Databricks has open-sourced “Omnigent” (Apache 2.0), a meta-harness positioned above existing AI Agent harnesses such as Claude Code and Codex. It enables composition, control, and sharing of multiple agents across different tools. Born from real-world usage inside Databricks with over 5,000 engineers, it is immediately usable by anyone.

📑Table of Contents

1. What is Omnigent — Databricks’ Meta-Harness Abstraction Layer for Agents
2. Core Features: Composition, Control, and Collaboration
3. Installation and Quick Start
4. Policy Settings and Security Architecture
5. Real-World Use Cases and Limitations
6. Frequently Asked Questions
7. Summary and Future Outlook

In my workflow I typically use Claude Code for architecture and Codex for code generation. The context fragmentation between agents has been a constant pain point. Omnigent solves this by providing a unified orchestration layer on top of multiple agents.

---

Core Features: Composition, Control, and Collaboration

Omnigent is built around three pillars.

Composition: Dynamic Multi-Agent Synthesis

Combine Claude Code’s strength in long-running tasks with Codex’s fast code generation inside a single session. Roles can be defined via policies and switched dynamically as needed.

---

Control: Centralized Policy Management

Set cost caps, token budgets, and runtime limits from one place. API keys and MCP connections are also unified through Omnigent.

---

Collaboration: Real-time Sharing and Team Workflows

Share sessions with teammates and review execution history or intermediate artifacts in real time. Native GitHub integration is included.

---

Installation and Quick Start

Clone the official repository and install with pip.

git clone https://github.com/omnigent-ai/omnigent.git
cd omnigent
pip install -e .

During first launch you can link with a Databricks account. API keys are managed via environment variables.

---

Policy Settings and Security Architecture

Omnigent places strong emphasis on security. The following policies are supported:

Policy	Example	Effect
max_tokens_per_session	50000	Token cap per session
allowed_models	claude-opus-4.7,codex-pro	Model whitelist
cost_limit_usd	10.0	Daily USD cap

Source: Databricks official blog (June 2026)

---

Real-World Use Cases and Limitations

Limitations: As an Alpha release, compatibility with some MCP tools remains incomplete. When running locally without the cloud sandbox, the scope of enforced security policies is narrower, so caution is advised.

---

Frequently Asked Questions

Q: How is Omnigent different from Claude Code or Codex?

Omnigent sits above them as a meta-harness. It does not replace individual agents; it orchestrates multiple agents through a single control plane.

Q: What are the prerequisites for installation?

Python 3.10+, plus API keys for each agent you intend to use. A Databricks account is optional but recommended.

Q: Can I set cost limits via policy?

Yes. Settings such as max_cost_per_day allow flexible daily USD caps.

Q: How do I share a session with my team?

Use the “share session” command from the CLI or Web UI to generate an invite link. GitHub team integration is also supported.

Q: Is the cloud sandbox mandatory?

Not mandatory, but recommended for production workloads due to security and scalability benefits.

Q: How stable is the Alpha release?

It has been used internally by 5,000+ Databricks engineers. External users still report bugs actively. Apply to mission-critical workloads with caution.

---

Summary and Future Outlook

Omnigent offers a practical answer to the long-standing “fragmentation” problem in AI agent usage. For developers already juggling Claude Code and Codex, the adoption barrier is low and the immediate benefit is high.

Further integrations with additional agent harnesses are expected. I will continue testing and will share updates as the project evolves. Being open source also means community-driven improvements are likely.

Furthermore, being open source, Omnigent invites community extensions. With insights from Databricks’ 5,000+ engineer organization, it has the potential to become a standard in future agent development workflows.

Related articles: Claude Opus 4.8 リリース：Claude CodeのDynamic Workflowsと高速・低コスト化を解説、Anthropic、Claude Fable 5 / Mythos 5 をリリース — 長時間・複雑タスク向け新世代モデル、Anthropic Fable 5 / Mythos 5、米政府輸出規制で全ユーザーアクセス停止 — Agent利用者に大混乱。

Related new article:

• Google Agent Development Kit (ADK) Open Source Release — Production-Grade Multi-Agent Framework – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Introducing Claude Tag: Make Claude a Proactive Team Member in Slack Channels – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Anthropic Adds Routines to Claude Code — Schedule, API & GitHub Triggers for 24/7 Autonomous Execution – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Anthropic’s Accusation Against Alibaba Reveals Growing AI Model Distillation Threats – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Claude in Microsoft Foundry GA: Build Production Agentic AI on Azure with NVIDIA GB300 – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Claude Fable 5 Export Controls Lifted: Global Redeployment Impact for Developers – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• From Prompts to Loop Engineering: Why Agent Topology Changes Developer Workflows – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Claude Fable 5 Official Prompt Engineering Guide: Free Illustrated Book and Practical Usage Tips – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Anthropic’s 4 AI Coding Loop Types: Stop Conditions and Practical Developer Workflows – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Fable AI Practical Usage Guide 2026 – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Anthropic’s New Technique to Visualize Claude’s Internal Reasoning Process – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Alibaba Bans Claude Code for Employees Starting July 10 Over Alleged Backdoor Risks – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Anthropic launch-your-agent Claude Code Skill Open Source — Build Managed Agents Fast – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.
• Free Anthropic Academy courses for Claude Code, MCP, and Agent Skills – This published update adds current operational context for Omnigent: Databricks Open-Sources Meta-Harness for Multi-Agent Control Across Claude Code and Codex.