What are AI agent frameworks?

AI agent frameworks are libraries that handle the orchestration layer between language models and the rest of your application — managing tool calls, memory, multi-step reasoning loops, and agent-to-agent coordination. Without them, you'd write your own retry logic, prompt chaining, state management, and tool dispatch from scratch. This directory lists 23 frameworks ranging from lightweight single-agent libraries like PydanticAI and Smolagents to full multi-agent orchestration platforms like CrewAI, AutoGen, and MetaGPT.

How do I choose the right AI agent framework?

Four criteria matter most: (1) Single vs. multi-agent — if you're building one autonomous agent, PydanticAI, Instructor, or Claude Agent SDK are simpler; if you need agent teams with roles, use CrewAI, AutoGen, or Camel AI. (2) LLM lock-in — LangChain, LlamaIndex, and Haystack are model-agnostic; OpenAI Agents SDK and Google ADK are vendor-specific. (3) Persistence and memory — Letta and Mem0 are purpose-built for stateful agents with long-term memory, while most others treat memory as secondary. (4) Production readiness vs. research — Rasa and Semantic Kernel are battle-tested in enterprise; BabyAGI and MetaGPT are better suited for experimentation.

What is the difference between LangChain and LlamaIndex?

LangChain is a general-purpose orchestration framework for building chains, agents, and pipelines with any LLM — it's broad by design and covers everything from prompt templates to agent execution loops. LlamaIndex is specialized for retrieval-augmented generation (RAG): it provides optimized data connectors, indexing strategies, and query engines for getting LLMs to reason over your own data. In practice, LangChain is the better choice when you're building an agent that calls tools and APIs; LlamaIndex is the better choice when your core problem is search and retrieval over a document corpus. Many production systems use both.

Are there free and open-source AI agent frameworks?

Nearly all 23 frameworks listed here are open-source: LangChain, LlamaIndex, CrewAI, AutoGen, Haystack, DSPy, Phidata, Smolagents, PydanticAI, BabyAGI, Camel AI, MetaGPT, Rasa, Letta, Mem0, and Agno are all MIT or Apache 2.0 licensed and free to self-host. The exceptions are tools with commercial tiers: Composio, SuperAGI, and Instructor have free plans but charge for hosted or advanced features. Google ADK, OpenAI Agents SDK, and Claude Agent SDK are free libraries but designed to work with their respective paid API services.

What is the difference between an agent framework and a prompt chaining library?

Prompt chaining libraries (like early LangChain or Instructor) execute a fixed sequence of LLM calls you define at build time — the flow is deterministic. Agent frameworks go further: the LLM itself decides at runtime which tools to call, in what order, and when to stop, based on intermediate results. Frameworks like CrewAI, AutoGen, Google ADK, and Phidata implement this agentic loop with built-in tool dispatch, observation handling, and re-planning. DSPy sits in between — it compiles and optimizes prompt pipelines automatically, but isn't a runtime agent orchestrator in the same sense.

23 Best AI Agent Frameworks

A curated collection of the best agent frameworks provide structured infrastructure for building autonomous AI systems that can reason, plan, execute tasks, and coordinate with other agents using language models and external tools. They abstract orchestration complexity, memory management, and tool integration so developers can focus on defining agent behavior and workflows.

Agent frameworks are the foundation of production AI systems, enabling developers to build autonomous agents that reason through problems, use external tools, and coordinate across multiple specialized agents. They range from general-purpose orchestration platforms to specialized frameworks optimized for patterns like retrieval-augmented generation (RAG), conversational AI, multi-agent coordination, or code execution. The choice of framework depends heavily on your tech stack, deployment model, data privacy requirements, and the specific orchestration pattern your use case demands.

How to Choose

Evaluate agent frameworks across these dimensions:

Language and ecosystem compatibility — Python dominates, but C#/.NET (Semantic Kernel), Java (Semantic Kernel), and JavaScript options exist. Consider whether you need to integrate with existing backend systems.
Orchestration pattern — Simple sequential chains, multi-agent teams with role definitions, task-based workflows, conversation trees, or code-executing agents? Framework design differs significantly.
Memory and state management — Does your agent need persistent memory across sessions (Letta, Mem0), or is stateless request-response sufficient?
Deployment constraints — Cloud-managed platforms, self-hosted open-source, or fully on-premise for regulated industries? Data privacy may mandate specific choices.
Tool integration breadth — Do you need 250+ pre-built connectors (Composio) or will custom integrations work?
Data handling at scale — RAG systems handling thousands of documents require frameworks specifically optimized for retrieval and indexing (LlamaIndex, Haystack).
Observability and debugging — For production systems, built-in tracing (OpenAI Agents SDK, PydanticAI) versus add-on observability matters.

Comparison

Name	Best For	Pricing	Key Differentiator
LangChain	Production LLM applications with complex agent pipelines	See website	Largest ecosystem; most mature community integrations and examples
CrewAI	Multi-agent automation and knowledge work automation	Open-source + CrewAI AMP enterprise	Visual workflow editor; AI copilot for non-technical workflow builders
AutoGen	Multi-agent systems requiring fine-grained conversation control	Open-source (Microsoft)	Diverse conversation patterns; strong for distributed multi-agent coordination
Claude Agent SDK	Production agents leveraging Anthropic's models	Free SDK + token-based API costs	Native Claude integration; optimized for extended reasoning and complex workflows
LlamaIndex	RAG systems and document-centric AI applications	Free + LlamaCloud managed tier	Purpose-built data indexing; excellent document handling and retrieval quality
Haystack	Production RAG pipelines in regulated industries	Open-source + enterprise support available	Full pipeline visibility; auditable and serializable workflows for compliance
PydanticAI	Type-safe agents with structured, validated outputs	Open-source	First-class validation; integrates naturally with FastAPI and Pydantic ecosystems
OpenAI Agents SDK	Multi-agent workflows with task handoffs and human-in-the-loop	Free SDK + token-based API costs	Built-in tracing and guardrails; native voice/realtime agent support
Semantic Kernel	Enterprise AI in .NET, Python, or Java environments	Open-source (Microsoft)	Deep Azure integration; designed for complex enterprise system integration
Letta	Personal AI assistants with long-term user memory	Free self-hosted + managed platform	Transparent, editable agent memory; extends effective context window indefinitely
Agno	High-performance production agents with privacy requirements	Open-source framework + AgentOS paid runtime	Self-hosted production control plane; strong security posture for sensitive workloads
Smolagents	Minimal, transparent code-executing agents	Open-source	Lightweight foundation; ideal for data analysis and custom agentic workflows
Composio	Agents needing authentication-aware tool integrations	See website	250+ pre-integrated tools with OAuth; eliminates custom API integration work

Rasa

Open-source conversational AI framework

Frameworks

Production-grade framework for building text and voice assistants. On-premise deployment for data privacy.

Camel AI

Communicative agents for mind exploration

Frameworks

Multi-agent framework using role-playing for autonomous cooperation between AI agents. Research-oriented.

BabyAGI

AI-powered task management system

Frameworks

Lightweight autonomous agent that creates, prioritizes, and executes tasks using LLMs.

SuperAGI

Open-source autonomous AI agent framework

Frameworks

Developer-first framework for building, managing, and running autonomous AI agents with a marketplace of tools.

Instructor

Structured outputs from LLMs

Frameworks

Library for getting structured, validated data from LLMs. Built on Pydantic for type-safe agent outputs.

Mem0

Memory layer for AI agents

Frameworks

Add persistent memory to AI agents. User, session, and organizational memory for personalized experiences.

Letta

Stateful LLM agents with long-term memory

Frameworks

Framework for building agents with persistent memory. Formerly MemGPT. Unlimited context via memory management.

DSPy

Programming—not prompting—language models

Frameworks

Stanford framework for algorithmically optimizing LM prompts and weights. Moves beyond manual prompt engineering.

CrewAI

Framework for orchestrating role-playing AI agents

Frameworks

Multi-agent framework where AI agents work together as a crew, each with defined roles and goals.

LangChain

Build context-aware, reasoning applications

Frameworks

The most popular framework for building LLM-powered applications with chains, agents, and retrieval.

PydanticAI

Agent framework from the Pydantic team

Frameworks

Type-safe AI agent framework built on Pydantic. Strong validation and structured outputs.

Semantic Kernel

Microsoft's SDK for integrating AI into applications

Frameworks

Lightweight SDK from Microsoft for integrating LLMs into C#, Python, and Java applications.

Phidata

Build AI assistants with memory, knowledge, and tools

Frameworks

Framework for building autonomous AI assistants with long-term memory and tool use capabilities.

Google ADK

Google's Agent Development Kit

Frameworks

Google's framework for building AI agents with Gemini models. Supports multi-agent orchestration.

Smolagents

Hugging Face's lightweight agent framework

Frameworks

Minimalist agent library from Hugging Face for building code-executing AI agents.

Claude Agent SDK

Build production agents with Claude

Frameworks

Anthropic's official SDK for building AI agents powered by Claude models.

Agno

Build lightning-fast multi-modal agents

Frameworks

High-performance framework for building multi-modal AI agents with built-in memory and tool use.

AutoGen

Microsoft's framework for multi-agent conversations

Frameworks

Framework for building LLM applications via multi-agent conversation. Supports diverse conversation patterns.

Composio

Integration platform for AI agents

Frameworks

250+ tool integrations for AI agents. Connect agents to GitHub, Slack, Salesforce, and more.

LlamaIndex

Data framework for LLM applications

Frameworks

Connect custom data sources to large language models. Best for RAG and knowledge-intensive agent applications.

OpenAI Agents SDK

Build agentic AI apps with OpenAI

Frameworks

OpenAI's framework for building multi-agent workflows with handoffs, guardrails, and tracing.

MetaGPT

Multi-agent framework assigning different roles to GPTs

Frameworks

Assigns different roles (product manager, architect, engineer) to AI agents for collaborative software development.

Haystack

End-to-end NLP framework for building search and RAG

Frameworks

Open-source framework by deepset for building production-ready NLP and RAG pipelines.

Top Frameworks Experts

Are you an expert working with frameworks tools? Get listed and reach companies looking for help.

Claim Your Profile Get Verified

Frequently Asked Questions

Rasa

Open-source conversational AI framework

Frameworks

Production-grade framework for building text and voice assistants. On-premise deployment for data privacy.

Camel AI

Communicative agents for mind exploration

Frameworks

Multi-agent framework using role-playing for autonomous cooperation between AI agents. Research-oriented.

BabyAGI

AI-powered task management system

Frameworks

Lightweight autonomous agent that creates, prioritizes, and executes tasks using LLMs.

SuperAGI

Open-source autonomous AI agent framework

Frameworks

Developer-first framework for building, managing, and running autonomous AI agents with a marketplace of tools.

Instructor

Structured outputs from LLMs

Frameworks

Library for getting structured, validated data from LLMs. Built on Pydantic for type-safe agent outputs.

Mem0

Memory layer for AI agents

Frameworks

Add persistent memory to AI agents. User, session, and organizational memory for personalized experiences.

Letta

Stateful LLM agents with long-term memory

Frameworks

Framework for building agents with persistent memory. Formerly MemGPT. Unlimited context via memory management.

DSPy

Programming—not prompting—language models

Frameworks

Stanford framework for algorithmically optimizing LM prompts and weights. Moves beyond manual prompt engineering.

CrewAI

Framework for orchestrating role-playing AI agents

Frameworks

Multi-agent framework where AI agents work together as a crew, each with defined roles and goals.

LangChain

Build context-aware, reasoning applications

Frameworks

The most popular framework for building LLM-powered applications with chains, agents, and retrieval.

PydanticAI

Agent framework from the Pydantic team

Frameworks

Type-safe AI agent framework built on Pydantic. Strong validation and structured outputs.

Semantic Kernel

Microsoft's SDK for integrating AI into applications

Frameworks

Lightweight SDK from Microsoft for integrating LLMs into C#, Python, and Java applications.

Phidata

Build AI assistants with memory, knowledge, and tools

Frameworks

Framework for building autonomous AI assistants with long-term memory and tool use capabilities.

Google ADK

Google's Agent Development Kit

Frameworks

Google's framework for building AI agents with Gemini models. Supports multi-agent orchestration.

Smolagents

Hugging Face's lightweight agent framework

Frameworks

Minimalist agent library from Hugging Face for building code-executing AI agents.

Claude Agent SDK

Build production agents with Claude

Frameworks

Anthropic's official SDK for building AI agents powered by Claude models.

Agno

Build lightning-fast multi-modal agents

Frameworks

High-performance framework for building multi-modal AI agents with built-in memory and tool use.

AutoGen

Microsoft's framework for multi-agent conversations

Frameworks

Framework for building LLM applications via multi-agent conversation. Supports diverse conversation patterns.

Composio

Integration platform for AI agents

Frameworks

250+ tool integrations for AI agents. Connect agents to GitHub, Slack, Salesforce, and more.

LlamaIndex

Data framework for LLM applications

Frameworks

Connect custom data sources to large language models. Best for RAG and knowledge-intensive agent applications.

OpenAI Agents SDK

Build agentic AI apps with OpenAI

Frameworks

OpenAI's framework for building multi-agent workflows with handoffs, guardrails, and tracing.

MetaGPT

Multi-agent framework assigning different roles to GPTs

Frameworks

Assigns different roles (product manager, architect, engineer) to AI agents for collaborative software development.

Haystack

End-to-end NLP framework for building search and RAG

Frameworks

Open-source framework by deepset for building production-ready NLP and RAG pipelines.