Prompt Flow is Microsoft’s open-source toolkit for the full LLM application lifecycle: build flows that chain prompts, Python code and tools, tune variants systematically, evaluate quality with built-in metrics, and deploy — all with a visual designer in VS Code. It has earned about 11,300 stars on GitHub.
Prompt engineering by trial and error in a notebook does not scale to production applications. Prompt Flow treats an LLM application as a flow: a directed graph of nodes — LLM calls with templated prompts, Python functions, and external tools — that can be edited in code or in a visual canvas inside VS Code, with connections to model providers and your own data managed as reusable resources. Around that core, the toolkit covers the work that separates a demo from a product: run the same flow over batches of test data, create and compare prompt variants side by side, score outputs with built-in evaluators for groundedness, relevance and safety (or your own custom metrics), and trace every execution for debugging. Prompts and flows are files in your repository, so they version with your code and flow through normal CI. It integrates with Azure AI services when you want cloud deployment and monitoring, but the toolkit itself is fully open and runs locally. Development continues actively in 2026.
Prompt Flow is free and open source under the MIT license. Optional Azure AI integration is billed under Azure pricing if you choose it, but the toolkit itself costs nothing and works locally.
Prompt Flow is for developers and teams building LLM applications beyond the prototype stage — RAG systems, copilots and workflow automation — who need structured development, evaluation and iteration on prompts and flows rather than ad-hoc experimentation.