Z80-μLM is a conversational language model trained via quantization-aware training to run within 64KB RAM on vintage 1976-era Z80 processors. The project includes tools for training models in Python, exporting them as executable CP/M binaries, and includes pre-built examples like tinychat (a conversational chatbot) and guess (a 20 Questions game). Training data generation tools can leverage LLMs via Ollama or Claude API.