* Automated planning and self-evolution, so your AI “thinks before it acts” and learns from mistakes * Extended tool usage, including web browsing, search engine queries, and LLM-driven text/image analysis * [Experimental] Multi-agent collaboration, with o1 doing the planning, and regular Claude/GPT-4o doing the execution.