A working map of the platforms teams reach for to build, orchestrate, and operate AI agents — from graph-based orchestration and multi-agent crews to retrieval, observability, and guardrails.
Building an AI agent rarely means writing one program end to end. Most teams reach for a graph or chain framework to define the workflow, a multi-agent layer to split work across specialised roles, a retrieval layer to ground answers in real data, and an observability tool to see what the agent actually did in production. This directory groups 23 frameworks and platforms by the job they do in that stack.
Visualize and orchestrate AI workflows elegantly with its graph-based interface. The langgraph package on PyPI was at 1.2.16 (16 October 2026).
Team up multiple AI agents to perform collaborative problem-solving. CrewAI 1.15.28 was released on 14 October 2026, according to GitHub.
Straightforward library for building agentic applications where a model can use additional context and tools, hand off to other specialised agents, stream partial results, and keep a full trace. The openai-agents-python repository's v0.24.0 was released on 15 October 2026.
Microsoft Research framework for building multi-agent applications. Microsoft Agent Framework, which builds on AutoGen and Semantic Kernel, is the newer option. The AutoGen README says the project is in maintenance mode, limited to bug fixes, security patches and documentation, and recommends Microsoft Agent Framework for new projects. The latest GitHub release listed is python-v0.7.5, from September 2025.
Platform of AI-native apps and digital workers for sales, marketing and support. It began as an open-source autonomous agent framework. The latest GitHub release listed is v0.0.14, from January 2024.
Bring together search, reasoning, and AI to provide intelligent agent capabilities. Haystack v3.4.1 was released on 21 October 2026, according to GitHub.
Connecting LLMs with structured and unstructured data sources for intelligent data retrieval — the go-to framework for building knowledge-augmented AI applications. LlamaIndex v0.14.25 was released on 21 September 2026, according to GitHub.
Open-source development kit for building AI agents and integrating AI models into C#, Python or Java code. Microsoft Agent Framework builds on it and AutoGen. Semantic Kernel .NET 1.81.0 was released on 6 October 2026, according to GitHub.
Popular for chaining AI components, enabling complex agent workflows. The langchain package on PyPI was at 1.4.7 (21 October 2026).
Now an autonomous security product: security workers that defend repositories and agent-built software. It began as an open-source agent framework.
Simplify deployment and management of AI agents at scale. The latest GitHub release listed is 0.4.21, from August 2025.
Build, test, and refine AI agents with a comprehensive toolset. The latest GitHub release listed is v0.5.1, from February 2025, though the repository was updated in August 2026.
A studio environment for rapid prototyping of AI agent applications.
Leverage large language models to create agents that manage real-world projects. The MetaGPT repository now lives under FoundationAgents/MetaGPT, and its About section links atoms.dev; the latest GitHub release listed is v0.8.1, from April 2024, though the repository was updated in January 2026.
A framework for programming, rather than prompting, language models, with optimizers that tune prompts and weights for agents and pipelines. DSPy 3.4.1 was released on 12 October 2026, according to GitHub.
Microsoft's open-source framework for building, orchestrating and deploying agents and multi-agent workflows in Python and .NET, building on AutoGen and Semantic Kernel. The python-1.22.0 release was published on 15 October 2026, according to GitHub.
Open-source, code-first toolkit from Google for building, evaluating and deploying agents and multi-agent systems. The adk-python v2.12.0 release was published on 13 October 2026, according to GitHub.
Anthropic's SDK for building agents that use the same tools, agent loop and context management as Claude Code. The claude-agent-sdk-python v0.2.172 release was published on 22 October 2026, according to GitHub.
Python agent framework from the Pydantic team, focused on type-safe, structured outputs and model-agnostic agent development. Pydantic AI v2.59.0 was released on 21 October 2026, according to GitHub.
TypeScript framework for building AI agents and workflows, with built-in memory, tools and evaluation. @mastra/core 1.79.0 was released on 21 October 2026, according to GitHub.
Open-source SDK from AWS for building AI agents with a model-driven approach and a few lines of code. The strands-agents package on PyPI was at 1.62.0 (21 October 2026).
Hugging Face's minimal library for building agents that write and run their actions as code. smolagents v1.26.0 was released on 29 May 2026, according to GitHub.
Framework and runtime for building multi-agent systems with memory, knowledge and tools. Agno v3.1.8 was released on 21 October 2026, according to GitHub.
No frameworks match that search. Try a different term or clear the filter.
An AI agent framework is a software library or platform that provides the building blocks for agents: a loop that lets a model plan and call tools, memory and state, orchestration of multi-step or multi-agent workflows, and often tracing and evaluation. It saves teams from writing that plumbing from scratch.
Orchestration frameworks such as LangGraph and LangChain define the workflow around the model. Multi-agent frameworks such as CrewAI, AutoGen and Agno coordinate several specialised agents. RAG frameworks such as LlamaIndex and Haystack focus on grounding agents in your data. Many frameworks cover more than one of these.
Start from your language and cloud: Microsoft Agent Framework suits .NET and Azure teams, Strands Agents suits AWS, Google ADK suits Google Cloud, Mastra suits TypeScript teams, and Pydantic AI suits Python teams that value typed outputs. For maximum flexibility and community, LangGraph and CrewAI are widely used. Prototype the same task in two frameworks before committing.
No. A simple agent can be a loop around a model API with a few tool functions, and vendor SDKs such as the OpenAI Agents SDK or Claude Agent SDK keep this light. Frameworks pay off when workflows branch, need persistent state or human approval, or span several agents.
Use an observability tool that traces each model call, tool call and decision, such as AgentOps, or the tracing built into your framework. Pair it with evaluation and guardrails, and see the Agentic AI observability and governance guides on this site.
Its README says AutoGen is in maintenance mode, with contributions limited to bug fixes, security patches and documentation, and recommends Microsoft Agent Framework for new projects. The latest AutoGen release listed on GitHub is python-v0.7.5 from September 2025, while Microsoft Agent Framework published python-1.22.0 on 15 October 2026.
According to GitHub and PyPI in October 2026: Claude Agent SDK (v0.2.172, 22 October), Pydantic AI (v2.59.0, 21 October), Agno (v3.1.8, 21 October), Mastra (@mastra/core 1.79.0, 21 October), Haystack (v3.4.1, 21 October), OpenAI Agents SDK (v0.24.0, 15 October), CrewAI (1.15.28, 14 October), Google ADK (v2.12.0, 13 October) and DSPy (3.4.1, 12 October). SuperAGI, AgentOps, AgentForge and MetaGPT show older latest releases, so check their repositories before adopting them.