YeQing17-2026/OmniAgent
An open-source agent framework that self-evolves its skills, context, and decision-making model during interaction using LLM-driven mechanisms.

OmniAgent is an agent framework implementing full-dimensional self-evolution through four core mechanisms: proactive memory using dual-path LLM alignment, automatic skill creation and repair during interaction, context personalization via multi-layer information stacks, and BrainModel evolution through online reinforcement learning feedback loops. The framework also includes Hyper Harness for safe execution scaffolding and Deep Reflexion for improving task success rates.