Building an AI agent with human features: persistent memory, mood, and evolving skills

In the age of AI agents that mimic human behavior, Pavel has developed his own framework called PaulusAI. This innovative creation stands out by incorporating long-term memory, emotional states, and the ability to learn new skills.

Unlike other popular AI frameworks like OpenClaw and Hermes Agent, PaulusAI aims to create a more companion-like experience. It achieves this through a persistently stored emotional state, which subtly influences its communication style. This mood system modulates the agent's tone and responses, making it feel more human and less like a simple utility.

PaulusAI's long-term memory is not based solely on a massive context window. Instead, it uses remember and recall tools to consciously store and retrieve facts about the user and the world. This allows the agent to maintain a sense of continuity and familiarity with the user over time.

The framework also emphasizes skill acquisition. When the agent encounters a new, non-trivial task, it can save a procedure for future use, effectively learning how to be more useful. This ongoing learning process ensures that the agent evolves and adapts to the user's needs.

Safety is an essential aspect of PaulusAI. While it can run commands, manage files, and interact with external gateways, it is designed to pause for human approval on high-impact actions. This ensures that the agent remains under human control and does not make critical decisions without proper oversight.

For those interested in exploring or contributing to this innovative project, the PaulusAI repository is open-source. The author, Pavel, is actively seeking contributors who can refine memory retrieval, expand emotional logic, or build new autonomous tools.

To demonstrate the effectiveness of PaulusAI, Pavel deployed the agent on his VPS and even used it to learn French. He finds the experience highly enjoyable and believes that the agent will enhance his life in various ways.

In a final note, Pavel shares his thoughts on the title image generated by the agent using the Gemma 4 31B model. He describes it as a physical manifestation of memory and connections, capturing the balance between logic and emotion. Pavel feels honored to have his creation represented this way in his blog, noting that it makes him feel "seen."