Prompt flow

Microsoft built Prompt flow to carry LLM applications from prototype to production as executable flows: directed graphs that wire prompts, Python functions, and model calls into units that can be run, tested, and versioned like ordinary code. The pip-installable suite, promptflow plus promptflow-tools on Python 3.9 to 3.11, includes a local test runner, batch execution over datasets, built-in and custom evaluation flows for scoring output quality, and tracing that records every model interaction for debugging. A VS Code extension adds a visual flow designer, and the CLI covers scaffolding, testing, and deployment paths that package flows as Docker images or serve them locally. The same flow format runs inside Azure AI Foundry, which drives much of its enterprise adoption, but nothing requires Azure and flows execute against any OpenAI-compatible endpoint, hosted or local. MIT licensed with about 11,000 GitHub stars, it remains one of the more complete open answers to systematic prompt testing, evaluation, and LLMOps workflow discipline.