FastFlowLM is an NPU-first runtime designed to run large language models and other AI workloads on AMD Ryzen AI processors. Now part of AMD, the project focuses on efficient, on-device inference, offering a compact runtime of approximately 17 MB and a familiar command-line interface for installing models, running them locally, and serving them through OpenAI-compatible APIs. It supports text generation, reasoning, mixture-of-experts models, vision, audio transcription, embeddings, and retrieval-augmented generation workflows.

Featured model options include Qwen, GPT-OSS, Gemma, Whisper, and Llama. FastFlowLM provides curated model recipes, installation guidance, technical documentation, and benchmarks based on Ryzen AI hardware. The site reports context windows up to 256,000 tokens and highlights local, offline operation for privacy-conscious use.

Developers can use the runtime with existing compatible tools, while benchmark information covers performance, power use, and hardware telemetry. FastFlowLM is available for Windows, with support and community resources offered through Discord and GitHub