Summary

Omnigent provides a meta-harness for orchestrating diverse AI agents, including commercial offerings like Claude Code and Codex, alongside custom agents. It addresses the complexity of managing multiple AI tools by offering a unified interface, enabling seamless swapping or combination of agent harnesses. This empowers users to supervise, collaborate on, and govern AI workflows across devices, significantly enhancing productivity for AI developers and researchers.

How It Works

The core innovation is Omnigent's meta-harness architecture, which abstracts away the underlying agent implementations. It allows users to integrate various AI models and custom agents through a common layer, facilitating dynamic switching or parallel use within a single session. Key design choices include real-time, cross-device synchronization of sessions (terminal, browser, mobile), enabling collaborative co-driving and shared live debugging. Agents can be executed in ephemeral cloud sandboxes (Modal, Daytona) for isolated, scalable operation, and their actions are governed by a flexible policy engine.

Quick Start & Requirements

Installation is streamlined via a curl script or package managers (uv, pip, Homebrew). Primary requirements include Python 3.12+, uv, git, Node.js 22 LTS+, and tmux. Optional Databricks integration requires omnigent[databricks] and the Databricks CLI. The project offers a macOS desktop app and detailed deployment guides for server setups. Links to relevant installation guides are provided within the README.

Highlighted Details

• Unified Session Management: Synchronized sessions across terminal, web UI, and mobile devices.
• Multi-Agent Orchestration: Enables complex workflows by running and supervising multiple agents concurrently, facilitating delegation and review.
• Flexible Model Integration: Supports direct API keys, subscriptions via official CLIs, and any OpenAI/Anthropic-compatible gateways.
• Cloud-Native Execution: Agents can run in sandboxed cloud environments (Modal, Daytona) for scalability and isolation.
• Action Governance: A policy engine allows fine-grained control over agent actions, including approval workflows, spend caps, and tool restrictions.

Maintenance & Community

The README does not detail specific community channels (e.g., Discord, Slack) or notable contributors/sponsorships. Contribution guidelines are available in CONTRIBUTING.md.

Licensing & Compatibility

Licensed under the permissive Apache 2.0 license, permitting commercial use and integration into closed-source projects without significant restrictions beyond standard attribution.

Limitations & Caveats

The project is currently in an alpha state, indicating potential instability or ongoing development. Setup requires specific versions of several dependencies, although an installer script aims to simplify this. Cloud sandbox support is limited to Modal and Daytona, with expansion planned.