omnigent: an open-source meta-harness for orchestrating and governing multiple AI agents across devices

What it solves

Omnigent provides a unified orchestration layer (a "meta-harness") for various AI agents, eliminating the need to rewrite configurations when switching between different agent runtimes. It solves the problem of agent fragmentation by allowing users to combine multiple agents in a single session, manage them across different devices (terminal, browser, mobile), and enforce security and spending policies across all of them.

How it works

Omnigent acts as a wrapper around existing agents like Claude Code, Codex, Cursor, and others. It uses a server-client architecture where a local or cloud-deployed server manages sessions. Users can define custom agents via YAML files that specify prompts, tools (Python functions or MCP servers), and sub-agents. It supports various credential types, including API keys, subscriptions, and gateways (like OpenRouter or Ollama), and can run agents in isolated cloud sandboxes (e.g., Modal, E2B, Kubernetes) to ensure safety and security.

Who it’s for

It is designed for developers and teams who use multiple AI coding agents and want a single interface to manage, supervise, and collaborate on them. It is particularly useful for those who need to move between devices or require strict governance over agent actions and costs.

Highlights

• Multi-Agent Orchestration: Mix and match different agents (e.g., Claude Code and Codex) in one session to have them review each other's work or split tasks.
• Cross-Device Sync: Start a session in the terminal and continue it on a phone or browser via a web UI.
• Governance Policies: Implement rules to pause for approval before risky shell commands, cap token spend, or limit tool access.
• Collaborative Sessions: Share live sessions for real-time co-driving or fork conversations to work independently.
• Cloud Sandboxing: Run agents in disposable remote environments to avoid running untrusted code on a local machine.