Project Golem is an OS-level AI work system that combines language model capabilities with operational actions. It provides a dashboard for real-time dialogue, skill and tool extensions via MCP, long-term memory management and retrieval, and multi-agent discussion with consensus aggregation. The system supports multiple LLM backends including Gemini Web (Browser-in-the-Loop), Ollama for local/private models, and LM Studio for OpenAI-compatible local APIs.