Show HN: AutoBot – live voice control for long-running AI work (github.com)

AutoBot, a new self-improving AI agent, has demonstrated significant advancements in managing complex, long-running workflows, achieving an 18.5% higher task completion rate compared to OpenAI's Sol Max and surpassing Anthropic's Claude Opus 5 Max in benchmarks like OSWorld 2.0. With a #1 ranking on the AssistantBench hidden-test leaderboard, AutoBot enhances the efficiency of AI by autonomously improving its own framework and enabling persistent memory functionalities, circumventing the limitations of traditional context windows. This innovation is crucial for the AI/ML community as it represents a leap towards more robust and adaptive AI agents capable of managing intricate projects over extended periods. The architecture allows for hierarchical memory storage, independent validation of task changes, and clean execution processes, all of which ensure that tasks can be resumed and validated without user intervention. Moreover, AutoBot includes privacy mechanisms tailored for various contexts—personal, professional, and shared—enabling users to manage their data more securely. These capabilities not only enhance the functionality of AI in real-world applications but also address concerns about data privacy and operational fidelity.