omnigent-ai/omnigent

Omnigent is an open-source AI agent framework and meta-harness: orchestrate Claude Code, Codex, Cursor, Pi, and custom agents — swap harnesses without rewriting, enforce policies and sandboxing, and collaborate in real time from any device.

What it solves

Omnigent provides a unified orchestration layer (a "meta-harness") for various AI agents, eliminating the need to rewrite code when switching between different agent runtimes. It solves the problem of agent fragmentation by allowing users to combine multiple agents—such as Claude Code, Codex, Cursor, and custom agents—within a single session, while providing centralized governance, sandboxing, and cross-device accessibility.

How it works

Omnigent acts as a server-client architecture that wraps existing agent harnesses. It supports multiple credential types (API keys, subscriptions, gateways, and Databricks) and can run agents in local environments or cloud sandboxes (e.g., Modal, E2B, Kubernetes). Users interact with agents via a CLI, a web UI, or a native macOS desktop app. The system uses YAML-based agent definitions to specify prompts, tools (Python functions, MCP servers), and sub-agent hierarchies.

Who it’s for

  • Developers who want to use multiple AI agents and orchestrate them to work together (e.g., using one agent to review another's work).
  • Teams who need a collaborative environment where they can share live sessions, co-drive agents on a single machine, or fork conversations.
  • Enterprise users who require strict governance via policies to cap spending, limit tool access, or require manual approval for risky shell commands.

Highlights

  • Cross-Device Sync: Start a session in the terminal and continue it on a mobile browser or desktop app.
  • Multi-Agent Supervision: Mix and match different agent vendors in one session to delegate tasks or perform cross-reviews.
  • Cloud Sandboxing: Launch agents in disposable cloud environments to avoid running untrusted code on a local machine.
  • Policy-Based Governance: Enforce server-wide or session-specific rules for spending caps and tool execution approvals.
  • Extensible Agent Creation: Define new agents using simple YAML files that can include local Python functions and MCP servers.