Summary

OmniAgent is an open-source framework for self-evolving AI agents, inspired by OpenClaw, designed to address the need for dynamically adaptable and secure AI systems. It targets developers and power users seeking advanced agent autonomy, offering real-time evolution of skills, context, and decision-making models, alongside robust safety mechanisms. The primary benefit is an agent whose intelligence and security continuously improve through interaction.

How It Works

OmniAgent's core innovation is "OmniEvolve," enabling full-dimensional self-evolution across skills, context, and the brain model. This is achieved through proactive memory mechanisms that use explicit feedback and LLM induction, real-time skill self-evolution via auto-creation, inspection, and repair, and context evolution driven by user interaction and LLM summarization for personalization. The BrainModel evolves dynamically via online reinforcement learning. Complementing this are Hyper Harness, an efficient execution scaffold supporting dynamic multi-agent and concurrent tool execution, and Deep Reflexion, a dual-layer reflective architecture for risk interception and failure analysis, enhancing task success rates.

Quick Start & Requirements

• Installation: pip install -e . followed by omniagent onboard for interactive setup.
• Prerequisites: Python 3.11+, an LLM API key (DeepSeek, OpenAI, Anthropic, Ollama, Gemini).
• Running: omniagent chat for CLI, omniagent serve for Web UI (http://127.0.0.1:18790).
• Docs: Official documentation is "on the way."

Highlighted Details

• Implements "full-dimensional self-evolution" (Skill, Context, BrainModel) in real-time during execution.
• Features a "Four-Layer Dynamic Security Scanning" system described as "unbypassable."
• Utilizes "Hyper-Harness" for efficient, safe, and intelligent execution, including dynamic multi-agent coordination and concurrent tool execution.
• Employs a "Deep Reflexion" dual-layer reflective architecture to significantly improve task success rates (PASS@1).

Maintenance & Community

• Roadmap: Near-term plans include enhancing the Proactive Memory System, implementing an Agent Plan-Mode, expanding channel connectors, and improving multi-agent collaboration.
• Contribution: Open to contributions via pull requests.

Licensing & Compatibility

• License: GPL-3.0.
• Compatibility: Requires any code referencing this project to also be open-sourced under the GPL-3.0 license, potentially restricting commercial use or integration into closed-source systems without careful consideration.

Limitations & Caveats

Official documentation is still under development ("on the way"). The GPL-3.0 license imposes strong copyleft requirements, necessitating open-sourcing derivative works. Some advanced features like "Agent Plan-Mode" and expanded channel support are listed on the roadmap and not yet implemented.