AI agent frameworks
Open-source toolkits for building your own agents. Mostly Python and TypeScript, all production-grade.
Agency Swarm
Multi-agent framework built on OpenAI Assistants API with role-based agents and structured communication
AgentScope
Visual-first multi-agent framework built for distributed Python workflows
Agno
High-performance Python framework for multi-modal agents and teams
Arize Phoenix
Open-source LLM observability and evaluation with native OpenTelemetry tracing and self-hosting
AutoGen
Microsoft's multi-agent conversation framework with role-based agents and tool use
Braintrust
Modern AI evaluation platform built for developers who treat LLM quality as a first-class engineering problem
Composio
Managed integrations and auth layer for AI agents. 1,000+ apps, OAuth handled
CrewAI
Role-based multi-agent orchestration for production workflows
Dify
Open-source LLMOps platform with visual workflow builder, RAG pipeline, and self-host or cloud options
DSPy
Stanford's framework for programming LLMs with optimizable modules instead of hand-written prompts
Eko
Production-ready TypeScript framework for browser-native agent workflows
Embedchain (Mem0)
Memory layer for AI agents and applications, evolved from the Embedchain RAG framework
Flowise
Open-source no-code builder for LLM apps, self-hostable with a TypeScript stack
Griptape
Python AI agent framework with off-prompt data handling, Pipelines, Workflows, and Griptape Cloud
Haystack
The pipeline-first AI framework built for European enterprises and production RAG
Helicone
Open-source LLM observability proxy that logs every API call with zero code changes
Humanloop
Enterprise prompt management and evaluation platform for teams shipping LLM applications to production
Instructor
The simplest path to structured, typed outputs from any LLM using Pydantic
LangChain
The original agent framework that defined the chains, agents, tools, memory pattern
Langflow
Drag-and-drop visual builder for LangChain flows, with an API and self-host option
Langfuse
Open-source LLM observability, evaluation, and prompt management for agent debugging and cost tracking
LangGraph
Build stateful, multi-actor LLM applications as graphs
LangSmith
LangChain's observability and evaluation platform for debugging, testing, and monitoring LLM applications
Letta
Memory-first agent framework for stateful, persistent AI agents
LiveKit Agents
Real-time voice and video AI agents built on WebRTC infrastructure
LlamaIndex
RAG-first framework for connecting LLMs to your data with first-class document processing
Mastra
TypeScript-first agent framework with workflows, RAG, evals, and Cloud deploy
OpenAI Swarm / Agents SDK
OpenAI's official framework for multi-agent handoffs and production agent workflows
Phidata
Python agent framework, now superseded by Agno
Pydantic AI
Type-safe Python agent framework from the Pydantic team
Semantic Kernel
Microsoft's enterprise-grade AI agent framework for C#, Python, and Java
smolagents
HuggingFace's minimal agent framework where the LLM writes Python, not JSON
Swarms
Enterprise multi-agent orchestration framework for building large-scale cooperative agent swarms
Vellum
The GUI-first platform for building, testing, and shipping production AI workflows
VoltAgent
TypeScript agent framework with built-in VoltOps observability for production AI systems