hexdocom/lemonai

Implements a Docker-based VM sandbox architecture for isolated code execution while supporting local LLMs (DeepSeek, Qwen, Llama) via Ollama/VLLM, ensuring zero cloud dependency with optional cloud model fallbacks. Features a novel HTML editor interface enabling iterative AI-assisted refinement of generated content without full regeneration, alongside persistent conversation memory for self-improvement across sessions. Executes web research, code generation, and data analysis tasks entirely locally with 10-100x lower operational costs compared to commercial agentic platforms.

1,472 stars.