Prompt flow

Prompt flow is an open-source suite of development tools from Microsoft, released under the MIT License, designed to streamline the full lifecycle of LLM-based AI application development. It covers ideation, prototyping, testing, evaluation, production deployment, and monitoring — all in one cohesive toolkit. The project is hosted on GitHub at microsoft/promptflow and has accumulated over 11,000 stars since its creation in June 2023.

What It Is

Prompt flow is a Python-based developer toolkit that lets teams build production-quality LLM applications by composing flows — executable pipelines that link LLMs, prompts, Python code, and other tools together. It is not a no-code platform; it targets developers and ML engineers who need repeatable, testable, and deployable AI workflows. The core library is installable via pip install promptflow promptflow-tools and requires Python 3.9–3.11.

Core Workflow

The development process in prompt flow follows three main stages:

• Flow development: Create and iterate on flows using a DAG-based YAML definition (flow.dag.yaml) that specifies inputs, outputs, nodes, connections, and LLM model configuration.
• Quality evaluation: Run batch evaluations against larger datasets and integrate testing into CI/CD pipelines to gate production readiness.
• Deployment: Deploy flows to a serving platform of choice or embed them directly into application code. An optional cloud-hosted version is available via Prompt flow in Azure AI (Azure Machine Learning).

Tooling and Interfaces

Prompt flow ships with multiple interfaces to fit different developer preferences:

• CLI (pf): A command-line interface for initializing flows, managing connections, running tests, and batch evaluations. Full reference documentation is published at the project docs site.
• VS Code Extension: A visual flow designer available on the Visual Studio Marketplace (published as prompt-flow.prompt-flow) that provides an interactive UI for building and debugging flows.
• Python SDK: Programmatic access to all flow operations for integration into existing Python codebases and automation scripts.
• GitHub Codespaces: A pre-built development environment is available for instant cloud-based onboarding directly from the repository.

LLM Tracing and Evaluation

A notable capability is built-in tracing of LLM interactions during flow execution, making it easier to debug multi-step prompting chains. The evaluation subsystem supports running flows against datasets and computing quality metrics, which can be wired into CI/CD systems. The project documentation includes an end-to-end tutorial covering prompt tuning, batch testing, and evaluation for production readiness.

Azure AI Integration

While the core toolkit is fully open-source and runs locally, Microsoft also offers a cloud-hosted version integrated into Azure AI (Azure Machine Learning). This optional path adds team collaboration features, managed compute, and enterprise-scale monitoring. The local open-source version and the Azure-hosted version share the same flow format, enabling a smooth transition from local development to cloud deployment.

Update: promptflow 1.17.1

The latest release is promptflow 1.17.1, published on January 9, 2025. The repository remains actively maintained, with the last code push recorded in June 2026 and 79 open issues at the time of data collection. The project's GitHub topics include llm, prompt-engineering, ai-application-development, and chatgpt, reflecting its positioning as a general-purpose LLM app development framework.