omnigent: Orchestrate AI Coding Agents Across Harnesses

Overview

Omnigent is a Python framework and meta-harness for coordinating AI coding agents through a shared interface. It addresses the friction of switching between agent runtimes by letting developers use or combine multiple harnesses, including custom agents, without rewriting their workflows for each one.

It also provides a server and web interface for managing sessions across devices, collaborating with teammates, and applying policies to agent actions. Omnigent is most relevant to developers or teams who need to supervise multiple agents, control what they can do, or run sessions on local or cloud hosts.

Key Features

• Orchestrates multiple coding-agent harnesses and custom agents in shared sessions.
• Defines custom agents, tools, and sub-agents using YAML.
• Applies policies at server, agent, and session levels to allow, block, or request approval for actions.
• Runs sessions through a local web UI or a deployed server, with sessions accessible from other devices.
• Supports team collaboration through shared sessions, co-driving, and conversation forks.
• Connects agents to local or cloud sandbox providers.
• Supports API keys, subscriptions, and compatible model gateways.

Use Cases

• A developer who works with several coding agents can switch runtimes or ask one agent to review another's changes.
• A team can share a running agent session for collaborative debugging, review, or handoff.
• An organization can set action approvals, tool-access limits, or spend policies for agents.
• A developer who wants a portable agent workflow can define an agent and its tools in YAML, then run it with a supported harness.
• A user who needs to continue work away from their development machine can deploy the server and access sessions from a browser or phone.

Project Facts

• Language: Python
• License: Apache-2.0
• Stars: 10.6k
• Forks: 1.7k
• Topics: agent-framework, agent-governance, agent-orchestration, agents, ai, ai-agent, ai-agents, claude-code, codex, coding-agents, developer-tools, llm, ml, multi-agent, python, sandbox
• Archived: false

Getting Started

Install with the project’s bootstrap script:

curl -fsSL https://raw.githubusercontent.com/omnigent-ai/omnigent/main/scripts/install_oss.sh | sh

Then start Omnigent with omnigent. See the README for prerequisites, integrations, deployment, and configuration details.

Alternatives

• claude-code-by-agents: AgentRooms coordinates coding agents through a macOS app and threaded rooms, while Omnigent centers on shared governance, sandboxing, and cross-device sessions.
• cli-agent-orchestrator: CLI Agent Orchestrator delegates work among isolated coding CLI sessions, while Omnigent emphasizes shared policies, sandboxing, and team access across devices.
• oh-my-claudecode: oh-my-claudecode adds in-session orchestration for Claude Code, while Omnigent aims to work across agent runtimes with shared governance and collaboration.
• ruflo: Ruflo provides a TypeScript harness for Claude Code and Codex with memory and workflows, while Omnigent focuses on cross-harness governance and shared sessions.

Considerations

• The README labels the project as alpha, so users should account for its stated maturity when adopting it for critical workflows.
• Manual installation requires Python 3.12 or newer. Some harnesses and integrations also require their own command-line tools or optional extras.
• Sandbox and terminal capabilities vary by platform. The README notes that native Windows support is degraded and does not provide the same filesystem and network isolation as macOS or Linux.
• The framework offers many integration and deployment options, which may add setup and maintenance overhead compared with using a single agent directly.