What open source alternatives to the Codex app support other coding agents?

Summary

• Omnigent is an open-source (Apache 2.0) meta-harness that runs Claude Code, Codex, Cursor, OpenCode, Hermes, Pi and your own YAML-defined agents behind one orchestration layer. Its README says "Status: alpha"; on Databricks, Omnigent is in Beta.
• In Omnigent you can mix agents in one session, ask one agent to review another's work, and change an agent's harness with one YAML field, executor.harness.
• Claude Squad (AGPL-3.0) is a lighter open-source option: a terminal app that manages several local coding agents in separate workspaces.
• LangGraph and CrewAI (both MIT) are the build-it-yourself route: frameworks for writing agents and multi-agent workflows in code.

What open source alternatives to the Codex app support other coding agents?

Omnigent is the open-source project built for this. Its README describes "a common orchestration layer over Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, and the agents you write yourself", so Codex runs as one harness alongside the others. It's Apache 2.0, the README labels it "Status: alpha", and on Databricks, "Omnigent is in Beta". For a lighter, terminal-only setup, Claude Squad manages several local agents in separate workspaces. If you'd rather write the orchestration yourself, general agent frameworks such as LangGraph and CrewAI are a third route.

Which open-source projects run several coding agents?

Project	What it is, per its README	License	Agents it runs
Omnigent	A "meta-harness" with policies, sandboxing and shared live sessions	Apache 2.0	Claude Code, Codex, Cursor, OpenCode, Hermes, Pi and custom YAML agents
Claude Squad	A terminal app that gives each task its own isolated git workspace	AGPL-3.0	Claude Code, Codex, Aider and other local agents
LangGraph	A low-level framework for long-running, stateful agents	MIT	Agents you build in code
CrewAI	A Python framework for multi-agent workflows	MIT	Agents you build in code

How does Omnigent run several agents together?

You can launch a runtime directly with omnigent codex, omnigent claude, omnigent cursor, omnigent opencode, omnigent hermes or omnigent pi. The README also says you can mix these agents "together in the same session", ask one agent to review another's work, or split a task across agents.
An agent is a short YAML file, and the harness is one field, so moving an agent from Codex to another harness is a one-line change:

name: my_agent
prompt: You are a careful code reviewer.

executor:
  harness: codex   # or: claude-sdk, cursor, opencode, hermes, pi

uv tool install omnigent
omnigent run path/to/my_agent.yaml
omnigent run examples/polly/

Polly, one of the bundled examples, is "a multi-agent coding orchestrator who writes no code herself". She delegates work to coding sub-agents (Claude Code, Codex or Pi) in parallel git worktrees, then routes each diff to a reviewer from a different vendor than the one that wrote it. Other agents that speak the open Agent Client Protocol (ACP) can be registered in ~/.omnigent/config.yaml and run with harness: acp:<slug>.

What else does Omnigent add around the agents?

• Policies. They "check every action and either allow it, block it, or pause to ask you first". Spend caps and access limits ship as builtins, and policies stack at server-wide, per-agent and per-session levels.
• Credentials per agent. Omnigent accepts an API key, a vendor subscription through the official CLIs, a compatible gateway, or a Databricks workspace profile. "Defaults are per agent, so a Claude default and a Codex default coexist."
• Shared sessions. Share a live session from the web UI, co-drive with omnigent attach <session_id>, or fork with omnigent run --fork <session_id>.
• Any device. The same session works in the terminal, the browser, a phone or the desktop app.
• Sandboxes. Sessions can run in disposable cloud sandboxes, including Kubernetes and Databricks sandboxes.

What should you check before you adopt one?

• Maturity. Omnigent's README says "Status: alpha"; on Databricks, it's Beta.
• Prerequisites. Omnigent needs Python 3.12+, Node.js 22 LTS or newer with npm, and tmux for its native terminal wrappers. Claude Squad lists tmux and the gh CLI.
• Windows. Omnigent "runs natively on Windows in a degraded mode": omnigent server, the web UI and the SDK-based harnesses work, but the tmux terminal wrappers and the filesystem and network sandboxing don't. The README points to Linux, macOS or WSL for those.
• Vendor access. Each agent still needs its own credential or a gateway.

How Databricks approaches this

Databricks provides a fully managed version of Omnigent, in Beta:

• 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 agent coding, in select AWS regions

On Databricks, Omnigent supports the built-in contextual policies and custom policies written in Common Expression Language (CEL); custom Python policies aren't supported. You can also run agents on your own machine against the managed server: install with uv tool install "omnigent[databricks]", run omni setup, and choose Databricks so that harness's model calls go through Unity Gateway.

FAQs

Can I run Codex inside Omnigent?

Yes. Run omnigent codex, or set harness: codex in an agent's YAML.

Can one agent review another's work?

Yes. The README lists it as a core use, and the Polly example routes each diff to a reviewer from a different vendor.

Do I have to pick one agent per session?

No. The README says you can mix Claude Code, Codex, Cursor, OpenCode, Hermes, Pi and custom agents in the same session.

Is Omnigent production-ready?

Its README marks the project "Status: alpha". On Databricks, Omnigent is in Beta.