MicrosoftPromptflow is a set of Microsoft development tools for building LLM applications through the whole cycle: prototyping, testing, evaluation, deployment, and monitoring. Feature development ended on April 20, 2026, and the service retires on April 20, 2027, when it becomes read-only. Microsoft recommends migrating Promptflow workloads to Microsoft Agent Framework.
You build executable flows that link LLMs, prompts, Python code, and other tools. A flow is described in a YAML definition of inputs, outputs, nodes, connections, and the model to use, and API keys are held in named connections for OpenAI or Azure OpenAI. You can test a flow interactively as a chatbot, trace its interactions with the LLM to debug it, and iterate.
For quality, you run flows in batches against larger datasets, evaluate them with metrics, and wire testing and evaluation into a CI/CD system. Finished flows deploy to a serving platform of your choice or embed in an application's code. A VS Code extension provides a visual flow designer, and a cloud version runs in Azure AI for team collaboration.
Features
- Executable flows: link LLMs, prompts, Python code, and tools into a single flow
- Tracing: trace interactions with LLMs to debug and iterate on flows
- Batch evaluation: measure flow quality and performance on larger datasets
- CI/CD integration: run testing and evaluation in your pipeline
- Connections: named connections for OpenAI and Azure OpenAI credentials
- Deployment: serve flows on a platform of your choice or embed them in an app
- VS Code designer: a visual flow designer extension
- Azure AI version: a cloud edition for team collaboration
