Is there an open source tool for orchestrating multiple agent harnesses?

Summary

• Yes. Omnigent is an open-source (Apache 2.0) meta-harness. Its README describes "a common orchestration layer over Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, and the agents you write yourself" and lists the project status as alpha.
• In an Omnigent agent file, each sub-agent picks its own executor.harness, so one supervisor can delegate to a Cursor coder and a Claude Code reviewer.
• LangGraph and CrewAI are MIT-licensed open-source frameworks for writing multi-agent applications in code. Databricks' docs treat a meta-harness such as Omnigent as a separate option.
• Databricks documents Omnigent as a meta-harness option for agents on Agent Bricks. The managed version of Omnigent on Databricks is in Beta.

Is there an open source tool for orchestrating multiple agent harnesses?

Yes. Omnigent (Apache 2.0, github.com/omnigent-ai/omnigent) is an open-source meta-harness. Its README says you can swap or combine harnesses without rewriting your agent, and it supports Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, and custom agents. The README marks the project "Status: alpha", so confirm its stage before you depend on it. If you want to write the coordination logic of a multi-agent application yourself, LangGraph and CrewAI are open-source frameworks built for that job.

What is an agent harness, and what does a meta-harness add?

The Omnigent docs define it directly: "A harness is the runtime that executes your agent loop." Databricks lists LangGraph and the OpenAI Agents SDK as framework examples for that layer. For the multi-harness case, its docs point to a meta-harness: "Composing multiple harnesses, such as coding agents, into one agent with Omnigent." In Omnigent you change one line to swap the harness or model, while your tools, prompts, skills, and policies stay the same.

How do the open-source options differ?

Tool	What the project says it is	License	How you orchestrate
Omnigent	An open-source meta-harness: a common orchestration layer over Claude Code, Codex, Cursor, and other agents	Apache 2.0	YAML agent files; a supervisor delegates to sub-agents, each with its own harness and model
LangGraph	A low-level orchestration framework for building, managing, and deploying long-running, stateful agents	MIT	Code that defines state, nodes, and edges
CrewAI	An open-source Python framework for building production-ready multi-agent workflows	MIT	Crews of role-based agents, plus event-driven Flows

How does Omnigent orchestrate several harnesses?

An Omnigent agent is a short YAML file that you run with omnigent run path/to/my_agent.yaml. The Agent YAML spec says each sub-agent picks its own executor.harness and model, "so an orchestrator can mix harnesses by role". The sketch below uses only field names from the README and the spec, with placeholders for model IDs.

name: my_agent
prompt: You supervise a coder and a reviewer.

executor:
  harness: claude-sdk

tools:
  coder:
    type: agent
    executor:
      harness: cursor
      model: <cursor-model-id>
  reviewer:
    type: agent
    description: Review proposed code changes.
    prompt: You are a careful code reviewer.
    executor:
      harness: claude-sdk
      model: <model-id>

The README also describes supervising several agents in one session: "Ask one agent to review another's work, or split a task across agents that are each good at different things."

How do LangGraph and CrewAI orchestrate multiple agents?

LangGraph models an application as a graph. Its docs define state as "a shared data structure that represents the current snapshot of your application", nodes as "functions that encode the logic of your agents", and edges as "functions that determine which Node to execute next based on the current state". Its README highlights durable execution and human-in-the-loop oversight, and says LangGraph "can be used without LangChain".
CrewAI offers two building blocks. Crews are "teams of AI agents with true autonomy and agency, working together to accomplish complex tasks through role-based collaboration". Flows are "production-ready, event-driven workflows that deliver precise control over complex automations".

How does Databricks fit with open-source orchestrators?

Databricks documents a deployed agent as three layers: your framework or harness, an agent server, and an agent runtime. Agent Bricks "doesn't require a specific framework", and Agent Runtime "hosts agents built with any framework or harness, stateful or stateless". The Agent Bricks CLI (Beta) scaffolds LangGraph and OpenAI Agents SDK projects, and agentbricks init --existing brings in an agent you already built with either one. MLflow Tracing "records each step that your agent takes", and you can store and govern traces at scale in Unity Catalog.
Databricks also provides a fully managed version of Omnigent. Check these points before you plan around it:

Topic	What the Databricks docs say
Stage	"Omnigent is in Beta"
Availability	Requires the Omnigent preview enabled for your workspace and a region that supports Unity Gateway
Server and identity	A Databricks-operated Omnigent server that integrates with your workspace's identity provider
Models	Model access through the Foundation Model APIs and Unity Gateway
Sandboxes	Databricks Sandboxes, available in select regions on AWS
Policies	Built-in contextual policies and custom CEL policies; custom Python policies are not supported

FAQs

Do I need LangChain to use LangGraph?

No. The LangGraph README says it is built by LangChain Inc, the creators of LangChain, but "can be used without LangChain".

Can one Omnigent session mix agents from different vendors?

Yes. The README says you can mix Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, and custom agents together in the same session.

Can I limit what the agents are allowed to do?

Omnigent policies can pause for your approval before risky actions, cap spend, or limit which tools an agent reaches. They apply to the whole server, one agent, or a single chat.

Which stage is Omnigent in?

The open-source README shows "Status: alpha". Databricks states that "Omnigent is in Beta" for its managed version. Confirm the current stage before you plan a production rollout.