Show HN: I compared 15 AI agent frameworks across 4 production stacks

A developer published a side-by-side comparison of 15 AI agent frameworks, filtering by use case (RAG, multi-agent, structured output) and language (Python, TypeScript, .NET), updated weekly. The comparison covers frameworks like LangGraph, CrewAI, AutoGen, and others, highlighting their control flow, state management, and licensing. This resource helps developers choose the right framework for production stacks.

AI Agent Frameworks, Compared Side-by-side comparison of 15 AI agent frameworks. Filter by use case RAG, multi-agent, structured output and language Python, TypeScript, .NET . Updated weekly. 12k+Python / TypeScriptcontrol: graphstate: explicit-graphMITmulti-step agenthuman-in-the-loopstateful workflowRAG agent /frameworks/langgraph LangGraph Graph-based control for stateful LLM agents with explicit transitions and human gates. 28k+Pythoncontrol: crewstate: message-passingMITrole-based agent teamtask delegationresearchcontent workflow /frameworks/crewai CrewAI Role-based crew orchestration with sequential or hierarchical task delegation. 35k+Python / .NETcontrol: conversationalstate: message-passingMITmulti-agent chatcode generation agentresearchtool use /frameworks/autogen AutoGen Conversational multi-agent framework from Microsoft Research with deep tool use. 6k+Pythoncontrol: code-firststate: explicit-graphMITtyped structured outputfunction callingproduction appvalidation /frameworks/pydantic-ai Pydantic AI Typed agents built on Pydantic validation. Production-first, structured output as first class. 8k+Python / TypeScriptcontrol: code-firststate: implicitMITOpenAI-first stackhandoff patternguardrailstracing /frameworks/openai-agents-sdk OpenAI Agents SDK OpenAI official SDK with handoffs, guardrails, and built-in tracing UI. 10k+TypeScriptcontrol: code-firststate: explicit-graphElastic License 2.0TypeScript stackNext.js integrationworkflow with stepseval /frameworks/mastra Mastra TypeScript-native framework with workflows, agents, RAG, and built-in eval. 13k+TypeScriptcontrol: code-firststate: implicitApache 2.0streaming UIchat interfaceNext.js appedge runtime /frameworks/vercel-ai-sdk Vercel AI SDK TypeScript SDK with streaming UI primitives and provider abstraction. 94k+Python / TypeScriptcontrol: code-firststate: implicitMITRAGprototypebroad integrationchain composition /frameworks/langchain LangChain The original LLM framework. Broad integration surface, mature ecosystem. 36k+Python / TypeScriptcontrol: code-firststate: implicitMITRAGindex managementdata agentknowledge base /frameworks/llamaindex LlamaIndex Data-first framework. Strongest for RAG and structured data agent. 17k+Pythoncontrol: code-firststate: explicit-graphApache 2.0production RAGpipeline compositionenterprise search /frameworks/haystack Haystack Pipeline-based framework from deepset. Production search and RAG focus. 53k+TypeScript / Pythoncontrol: low-codestate: explicit-graphCustom open-core no-code agentworkflow builderteam collaborationself-host /frameworks/dify Dify Open-source LLM app platform with visual workflow builder. Self-hostable. 32k+TypeScriptcontrol: low-codestate: explicit-graphApache 2.0drag-drop chain builderno-codeprototypeself-host /frameworks/flowise Flowise Drag-and-drop LangChain workflow builder. Open source, self-host first. 65k+TypeScriptcontrol: low-codestate: event-drivenSustainable Use Licenseworkflow automationAI agent nodeself-hostAPI integration /frameworks/n8n n8n Workflow automation with AI agent nodes. Strong integration library. 21k+.NET / Python / Javacontrol: code-firststate: implicitMIT.NET stackenterpriseskill orchestrationplugin /frameworks/semantic-kernel Semantic Kernel Microsoft enterprise SDK. .NET / Python / Java with plugin orchestration. 12k+Pythoncontrol: code-firststate: implicitApache 2.0minimal footprintcode agenttool useHF ecosystem /frameworks/smolagents smolagents Hugging Face minimal agent framework. Code-as-action approach, small footprint.