vegaml ships CLI, MCP server, MLX engine and more
vegaml added a CLI, HTTP and MCP servers, a Docker image, ONNX export and a torch-free MLX engine for Apple silicon, plus cascade inference and a calibrate function, all under Apache 2.0.
vegaml just grew a whole toolbelt. All shipped this week, all Apache 2.0: - CLI, HTTP server, MCP server, Docker image - ONNX export and a torch-free MLX engine for Apple silicon - pydantic models as question sheets - a cascade: the 0.8B answers what it can settle, the 4B gets the rest, and what neither can settle is handed to you instead of guessed - calibrate(): refit the probabilities on 30 of your own labels - vegaml-eval: accuracy, calibration error and conformal coverage on your data, not mine - LangChain + LangGraph: route a graph or gate an input at decision-model speed pip install vegaml. Docs for every surface: https://github.com/NandhaKishorM/vegaml