New AI Agent Architecture to fix LLM deviations and token costs (github.com)

A new AI agent architecture has been introduced, designed to enhance the efficiency and reliability of large language models (LLMs) by utilizing a deterministic state machine to manage workflow. This innovative structure allows the LLM to focus on reasoning and tool calls for individual steps, while the state machine oversees the overall process flow. As a result, this approach significantly reduces token costs and enhances the predictability of multi-step automation, addressing key limitations of prior reliance on LLMs for comprehensive control. This advancement has considerable implications for the AI/ML community, particularly in automating complex tasks with greater precision and lower resource consumption. The architecture facilitates the easy setup and management of workflows through a user-friendly command-line interface, supporting customization of models and API keys from various providers like Anthropic, OpenAI, and Gemini. Furthermore, it introduces interactive workflows, skills for repeatable processes, and the ability to seamlessly incorporate external tools, which could foster increased adoption and experimentation in AI automation across different sectors.