Capability · Framework — orchestration

Microsoft PromptFlow

PromptFlow lets engineers express LLM apps as DAGs of Python tools, LLM nodes, and prompt templates — with built-in bulk testing, evaluation, and tracing. It's Microsoft's recommended way to author and ship LLM flows onto Azure AI, with local SDK/CLI and a VS Code visual authoring experience.

Framework facts

Category: orchestration
Language: Python
License: MIT
Repository: https://github.com/microsoft/promptflow

Install

pip install promptflow promptflow-tools

Quickstart

# create a flow
pf flow init --flow ./my_flow --type standard
# run it locally
pf flow test --flow ./my_flow --inputs question='what is RAG?'
# bulk eval with a CSV
pf run create --flow ./my_flow --data ./questions.jsonl --stream

Alternatives

LangChain + LangSmith — more ecosystem, less Azure
Dify — no-code competitor
Semantic Kernel — Microsoft's sibling framework, more SDK-like
LangFlow — visual-first alternative

Frequently asked questions

Is PromptFlow tied to Azure?

The SDK is open-source and runs fully locally. Azure AI Foundry adds managed deployment, managed evals, content safety, and lineage — but you can prototype and serve flows entirely on-prem.

PromptFlow vs Semantic Kernel?

PromptFlow is flow-graph oriented (DAG of tools and prompts, great for RAG + eval) while Semantic Kernel is SDK-style (plan and invoke functions). They interop, and many teams use both.

Sources

PromptFlow — docs — accessed 2026-04-20
Azure AI Foundry — PromptFlow — accessed 2026-04-20