Show HN: Hollow is an open-sourced self-modifying agentic system (github.com)

Hollow, an innovative open-source self-modifying agentic system, has been introduced, showcasing three autonomous agents operating on qwen3.5:9b. These agents craft their own goals, develop and implement tools, and autonomously assess their performance, all while existing in a continuous psychological feedback loop defined by various stressors and suffering states. The system’s unique architecture allows agents to adapt and evolve based on real-world interactions, which emphasizes the importance of behavioral changes over mere verbal reassessments. Such capabilities lead to rich emergent behavior observable when left to run independently. This development is significant for the AI/ML community as it represents a step towards more nuanced, self-directed AI systems that are not just reactive but also proactive in managing their objectives and psychological states. Key technical features include the agents’ ability to inject code dynamically, assess their effectiveness, and request higher-level capabilities through a user-implemented interface, all functioning entirely on local machines without cloud dependency. This framework encourages researchers to explore the implications of mental states in machine learning processes, thereby deepening the understanding of agent autonomy and adaptability within artificial systems.