Skip to content

Tools & frameworks

Go deeper · 30 min · no code

Before this: Code quality pipeline · After this: Developer tools In depth: Resources

The AI ecosystem moves fast. This page provides a curated overview of the tools and frameworks most relevant to building enterprise AI solutions — organized into orchestration frameworks, AI platforms, and developer tools.

Each entry includes a brief description, who it is for, and a link to official documentation.


  • Orchestration Frameworks

    SDKs and libraries for building AI agents, chains, and multi-agent systems.

    Jump to section

  • AI Platforms

    Cloud services for hosting models, managing data, and building end-to-end AI solutions.

    Jump to section

  • Developer Tools

    Tools that help developers build, test, and deploy AI-powered applications.

    Jump to section


Orchestration frameworks

Orchestration frameworks provide the building blocks for connecting large language models to tools, data, and each other. They handle prompt management, memory, tool calling, and multi-step workflows so you can focus on the application logic.

Microsoft Agent Framework

Microsoft's production SDK for building agents on .NET and Python. Agent Framework 1.0 shipped on 2026-04-03, merging Semantic Kernel and AutoGen into a single framework — Semantic Kernel as the foundation layer, with graph-based multi-agent orchestration on top.

If you are on the Microsoft stack, this is the current answer. Both predecessors are covered below because you will still encounter them.

Best for: .NET or Python teams on Azure; enterprise agent work needing first-class C# support.

Agent Framework documentation

Six runnable C# samples

The samples/ directory has six single-file C# programs on Agent Framework 1.x against Azure OpenAI: a basic agent, streaming, sessions, a C# function as a tool, an MCP server, and a workflow. Each declares its own packages, so dotnet run 1-basicagent.cs is the whole setup, and they need the same three environment variables as the Python labs.

All six were run end to end rather than only compiled. The readme also lists the renames from the pre-1.0 preview, none of which fail with a helpful message.

Semantic Kernel and AutoGen — what happened to them

Semantic Kernel is superseded rather than deprecated. Microsoft committed to critical bug fixes and security patches for at least a year after Agent Framework's GA, so roughly through April 2027. Existing code keeps working; new agent work should start on Agent Framework, and a migration guide is published.

AutoGen is in maintenance mode — no new features, no new orchestration patterns. Its last release was v0.7.5 in September 2025, and the repository has had no feature commits since. It is still widely recommended in listicles and comparison posts written before the merger; those are out of date.

LangChain / LangGraph

LangChain AI | Python, JavaScript/TypeScript

LangChain is a popular open-source framework for building applications with LLMs. It provides composable building blocks — chains, tools, retrievers, memory — that snap together to create complex LLM workflows.

LangGraph extends LangChain with a stateful, graph-based orchestration layer. Where LangChain chains are linear, LangGraph models workflows as directed graphs with cycles, conditional branching, and persistent state — making it ideal for building agents and multi-step reasoning systems.

Who it is for: Python and JavaScript developers building LLM applications, especially those who need flexibility and a large ecosystem of integrations.

Key capabilities:

  • Composable chains, tools, and retrievers (LangChain)
  • Stateful graph-based orchestration with cycles and branching (LangGraph)
  • Extensive integrations with vector stores, APIs, and data sources
  • Built-in support for RAG, agents, and conversational memory
  • LangSmith for observability, tracing, and evaluation

LangChain Documentation | LangGraph Documentation


CrewAI

CrewAI | Python

CrewAI is a role-based multi-agent framework that models teams of AI agents working together toward a goal. Each agent has a defined role, backstory, and set of tools. A "crew" orchestrates the agents, assigning tasks and managing collaboration.

Who it is for: Developers who want to build multi-agent systems using an intuitive role-and-task mental model, with minimal boilerplate.

Key capabilities:

  • Role-based agent definitions with goals and backstories
  • Task assignment and delegation between agents
  • Sequential and hierarchical crew execution
  • Integration with LangChain tools and LLM providers
  • Simple, readable agent definitions

CrewAI Documentation


OpenAI Agents SDK

OpenAI | Python, TypeScript

OpenAI's supported framework for building agents: tool calling, handoffs between agents, guardrails, sessions and tracing, with a small surface area. It replaced Swarm, which was explicitly an experiment and is still recommended by out-of-date comparison posts.

Who it is for: teams already on the OpenAI API who want the vendor's own abstractions rather than a general-purpose framework.

Agents SDK documentation


Google Agent Development Kit

Google | Python, Java

Google's open-source agent framework, with multi-agent composition, a local development UI, and evaluation built in rather than bolted on. Deploys to Google Cloud but is not locked to it.

Who it is for: teams on Gemini, or anyone who wants evaluation treated as a first-class part of the framework.

ADK documentation


How to choose

The decision is usually made for you by three things, in this order:

  1. Language. .NET narrows it to Microsoft Agent Framework. Python opens everything.
  2. Where the model lives. Azure pulls toward Agent Framework and Foundry; Gemini toward ADK; OpenAI direct toward the Agents SDK. All of them can call any OpenAI-compatible endpoint, so this is a pull, not a constraint.
  3. Whether you need a graph. Branching, cycles, checkpoints and resumable state are LangGraph's reason to exist. If your flow is linear, a framework buys you less than it costs.

A fourth consideration that is easy to miss: you can build the loop yourself. The build path does exactly that in about thirty lines, and knowing what a framework is doing for you is the difference between choosing one and inheriting one.


AI platforms

AI platforms provide the cloud infrastructure for hosting models, managing data, and building production-grade AI solutions. These are the services your AI applications run on.

Azure AI Foundry

Microsoft | Cloud Platform

Azure AI Foundry (formerly Azure AI Studio, and the destination for what was branded Azure AI Services) is Microsoft's unified platform for building, evaluating, and deploying AI solutions on Azure. It brings together model catalog access, prompt engineering tools, evaluation frameworks, and deployment pipelines in a single interface.

Who it is for: Teams building AI solutions on Azure who need a centralized platform for the full AI development lifecycle.

Key capabilities:

  • Access to OpenAI, Meta, Mistral, and other models via the model catalog
  • Prompt playground for experimentation
  • Built-in evaluation and red-teaming tools
  • Deployment to managed endpoints
  • Integration with Azure AI Search, Content Safety, and other Azure services

Azure AI Foundry Documentation


Azure OpenAI in Foundry models

Microsoft | Cloud Service

Azure OpenAI provides access to OpenAI's models hosted on Azure infrastructure. It adds enterprise features — virtual network support, managed identity, content filtering, and regional data residency — that are critical for production deployments.

Who it is for: Organizations that want to use OpenAI models with enterprise-grade security, compliance, and support.

Key capabilities:

  • Current OpenAI chat and reasoning models, embeddings, image generation, speech — see the model catalogue for what is available now
  • Enterprise security (VNet, private endpoints, managed identity)
  • Built-in content filtering and abuse monitoring
  • Regional deployment options for data residency
  • Managed and provisioned deployment options

Azure OpenAI Service Documentation


Microsoft | Cloud Service

Azure AI Search (formerly Azure Cognitive Search) is a fully managed search service that supports both traditional keyword search and modern vector search. It is the most common retrieval layer for RAG solutions built on Azure, enabling AI applications to find and surface relevant information from large document collections.

Who it is for: Teams building RAG solutions, enterprise search, or any application that needs to retrieve relevant information from large datasets.

Key capabilities:

  • Hybrid search combining vector similarity and keyword matching
  • Integrated vectorization with Azure OpenAI embeddings
  • Semantic ranking for improved relevance
  • Built-in skillsets for document cracking, OCR, and enrichment
  • Index management and query APIs

Azure AI Search Documentation


Developer tools

Developer tools help you build, test, and ship AI-powered applications more effectively. These tools integrate into your existing development workflow.

GitHub Copilot

GitHub | IDE Extension

GitHub Copilot is an AI pair programmer that provides code suggestions, completions, and chat-based assistance directly in your IDE. It supports VS Code, Visual Studio, JetBrains IDEs, and more. Copilot helps developers write code faster, understand unfamiliar codebases, and automate repetitive tasks.

Who it is for: Any developer who wants AI-assisted coding in their editor.

Key capabilities:

  • Inline code completions and suggestions
  • Chat interface for questions, explanations, and refactoring
  • Multi-file context awareness
  • Support for virtually all programming languages
  • Workspace and terminal integration

GitHub Copilot Documentation


Azure AI Document Intelligence

Microsoft | Cloud Service

Azure AI Document Intelligence (formerly Form Recognizer) extracts text, key-value pairs, tables, and structure from documents using pre-built and custom models. It powers intelligent document processing (IDP) pipelines, turning unstructured documents into structured data.

Who it is for: Teams building document processing pipelines — invoices, receipts, contracts, forms, and any document that needs automated data extraction.

Key capabilities:

  • Pre-built models for invoices, receipts, IDs, and more
  • Custom models for domain-specific documents
  • Layout analysis for tables, figures, and structure
  • Batch processing for high-volume scenarios
  • REST API and SDKs for .NET, Python, Java, and JavaScript

Azure AI Document Intelligence Documentation


Prompt Flow

Microsoft | Development Tool

Prompt Flow is a visual development tool for building, evaluating, and deploying LLM-based workflows. It provides a graph-based interface where you connect prompts, tools, and logic into executable flows — then evaluate them systematically before deploying to production.

Who it is for: AI developers who want a structured, visual approach to building and testing LLM applications, with built-in evaluation and CI/CD integration.

Key capabilities:

  • Visual flow editor for LLM workflows
  • Built-in evaluation framework with metrics
  • Local development and cloud deployment
  • Integration with Azure AI Foundry
  • CI/CD pipeline support for automated testing and deployment

Prompt Flow Documentation


Comparison at a glance

Tool Category Primary Use Language / Platform
Microsoft Agent Framework Orchestration Agents, workflows, multi-agent — successor to SK and AutoGen .NET, Python
OpenAI Agents SDK Orchestration Agents, handoffs, guardrails, tracing Python, TypeScript
Google ADK Orchestration Multi-agent composition with built-in evaluation Python, Java
LangChain Orchestration LLM application building blocks Python, JS
LangGraph Orchestration Stateful agent orchestration Python, JS
CrewAI Orchestration Role-based multi-agent systems Python
Azure AI Foundry Platform Full AI development lifecycle Azure
Azure OpenAI Service Platform Enterprise OpenAI model hosting Azure
Azure AI Search Platform Vector + keyword search for RAG Azure
GitHub Copilot Developer Tool AI-assisted coding IDE Extension
Document Intelligence Developer Tool Document data extraction Azure
Prompt Flow Developer Tool Visual LLM workflow builder Azure, Local

Go deeper

  • Microsoft Agent Framework — the current Microsoft SDK, and the migration target from Semantic Kernel and AutoGen.
  • OpenAI Agents SDK — small surface area, handoffs as a first-class concept.
  • Google ADK — multi-agent composition with evaluation built in.
  • LangChain and LangGraph — the widest integration ecosystem, and the graph layer for anything with cycles.
  • CrewAI — role-based orchestration with very little boilerplate.
  • Azure AI Foundry — the platform the Microsoft stack deploys onto.
  • Official sources — primary documentation for every vendor on this page.