Agent Frameworks
20+ frameworks for building AI agents in TypeScript and Python
LangChain
Full-stack framework for building LLM-powered applications with composable chains, agents, memory, and retrieval.
Best for: Teams needing a batteries-included framework with broad LLM provider support and production tracing.
LangGraph
Library for building stateful, multi-agent applications as cyclic graphs with built-in persistence and human-in-the-loop support.
Best for: Complex multi-agent workflows that need state management, branching logic, and production deployment.
CrewAI
Role-based multi-agent orchestration framework where AI agents collaborate as a crew with defined roles, goals, and backstories.
Best for: Teams wanting an intuitive role-based approach to multi-agent systems with minimal boilerplate.
OpenAI Agents SDK
Official OpenAI framework for building agentic applications with handoffs, guardrails, and tracing built in.
Best for: OpenAI-centric teams wanting first-party agent primitives with handoffs and safety guardrails.
Anthropic SDK
Official Anthropic SDK for building agents with Claude, featuring tool use, multi-turn conversations, and extended thinking.
Best for: Teams building with Claude who want direct SDK access to tool use, extended thinking, and computer use capabilities.
Vercel AI SDK
Full-stack TypeScript toolkit for building AI applications with streaming, tool calling, multi-step agents, and structured output.
Best for: TypeScript/Next.js developers building full-stack AI apps with streaming UIs and multi-provider support.
AutoGen
Microsoft framework for building multi-agent conversational systems where agents can chat with each other to solve tasks.
Best for: Research teams and enterprises needing flexible multi-agent conversations with code execution capabilities.
Composio
Tool integration platform that gives AI agents access to 250+ external tools and services with managed auth and execution.
Best for: Teams that need agents to interact with real-world SaaS tools without building individual integrations.
Mastra
TypeScript-first agent framework with built-in workflows, RAG, integrations, and an evaluation system for AI applications.
Best for: TypeScript developers wanting an opinionated, batteries-included agent framework with workflows and RAG.
Pydantic AI
Type-safe Python agent framework built on Pydantic, bringing validation, structured outputs, and dependency injection to AI agents.
Best for: Python developers who value type safety and want Pydantic's validation power applied to agent development.
BeeAI Framework
Open-source framework for building production-ready AI agents in Python and TypeScript with tool use and memory.
Best for: Developers wanting a lightweight, IBM-backed agent framework with production observability.
smolagents
Minimal Python agent framework by Hugging Face focused on code-based agents that write and execute Python to solve tasks.
Best for: Developers wanting a simple, code-first agent framework that leverages the Hugging Face ecosystem.
Semantic Kernel
Microsoft enterprise AI orchestration SDK for integrating LLMs into conventional applications with plugins and planners.
Best for: Enterprise teams integrating AI into existing .NET or Python applications with governance requirements.
Haystack
Modular NLP framework for building production-ready RAG pipelines, search systems, and question-answering applications.
Best for: Teams building production RAG systems and search pipelines with a modular, composable architecture.
LlamaIndex
Data framework for connecting LLMs to external data sources with advanced indexing, retrieval, and query engine capabilities.
Best for: Teams building data-intensive LLM applications that need sophisticated retrieval and indexing over private data.
Agno
Full-stack AI agent framework (formerly PhiData) for building agents with memory, tools, knowledge bases, and multi-agent teams.
Best for: Python developers wanting a batteries-included agent framework with memory, knowledge, and team coordination out of the box.
Agency Swarm
Multi-agent orchestration framework for creating collaborative AI agent swarms with customizable roles and communication flows.
Best for: Developers building collaborative multi-agent systems with defined communication hierarchies and shared state.
Instructor
Lightweight library for extracting structured outputs from LLMs using Pydantic models, with retries, validation, and streaming support.
Best for: Teams that need reliable structured data extraction from LLMs with type-safe validation and retry logic.
DSPy
Stanford framework for programming - not prompting - language models using composable modules, optimizers, and automated prompt tuning.
Best for: Research teams and ML engineers who want to optimize LLM pipelines programmatically instead of hand-tuning prompts.
Outlines
Structured text generation library that guarantees LLM outputs match a given format using regex patterns, JSON schemas, or grammars.
Best for: Teams running local models that need guaranteed structured output conforming to schemas, regex, or grammars.