Show HN: Local Agent – Local AI agent (Nova) with evolving memory (github.com)

Local Agent, an innovative AI agent framework developed over a weekend, enables users to experiment with local-first AI functionalities, focusing on learning agent runtimes, tool safety, and memory management. Designed for exploration rather than production, Local Agent runs entirely on user machines and supports multiple LLM providers, including Ollama and OpenAI, while utilizing a tool-based architecture that allows extensibility. Notably, it features a Retrieval-Augmented Generation (RAG) mechanism for document ingestion and semantic search, alongside built-in safety measures like deny-by-default permissions. A standout feature is the "nova" identity, which incorporates an experimental memory management system that allows the agent to autonomously adapt its long-term memory based on user interactions. This capability allows it to evolve its identity and knowledge but comes with warnings about unpredictable behavior, making it unsuitable for sensitive tasks. With support for SQLite-based persistent conversations, full execution context tracking, and comprehensive logging, Local Agent presents a sophisticated yet accessible playground for the AI/ML community, inviting developers to explore new frontiers in interactive AI agents.