Durable agentseve
Build agents with durable sessions, sandboxed execution, approvals, and multiple delivery channels.
Official documentation ↗Vercel
Find your agent framework.
From lightweight tool-calling SDKs to durable workflows and multi-agent systems. Explore the frameworks behind modern AI applications, including Vercel’s eve.
Editorial selection · Sources reviewed · Our methodology17 of 17 frameworks and SDKs
Durable agentsAgent SDKsAI SDK
Build AI applications with model providers, streaming responses, and tool-calling agents.
Official documentation ↗Vercel
Agent SDKsMastra
Build agents and workflows with memory, retrieval, evaluation, and application integrations.
Official documentation ↗Mastra
Durable agentsLangGraph
Orchestrate stateful agent workflows with persistence, streaming, and human oversight.
Official documentation ↗LangChain
Agent SDKsLangChain
Build tool-using agents with model integrations and reusable application components.
Official documentation ↗LangChain
Agent SDKsOpenAI Agents SDK
Build agents that use tools, delegate work, and maintain context across steps.
Official documentation ↗OpenAI
Agent SDKsClaude Agent SDK
Build agents with the agent loop and tools that power Claude Code.
Official documentation ↗Anthropic
Multi-agent systemsAgent Development Kit
Develop and orchestrate agents with tools, workflows, and evaluation support.
Official documentation ↗Google
Multi-agent systemsMicrosoft Agent Framework
Build agents and orchestrated workflows with tools, state, and human participation.
Official documentation ↗Microsoft
Multi-agent systemsCrewAI
Coordinate teams of specialized agents and build structured workflows with crews and flows.
Official documentation ↗CrewAI
Agent SDKsPydantic AI
Build typed agents with validated outputs, dependency injection, and model integrations.
Official documentation ↗Pydantic
Multi-agent systemsAgno
Build agents, teams, and workflows with memory, knowledge, and tool integrations.
Data & retrievalLlamaIndex
Build agents and workflows that use your documents and connected data sources.
Official documentation ↗LlamaIndex
Data & retrievalHaystack
Combine retrieval pipelines, language models, and tool-using agents in AI applications.
Official documentation ↗deepset
Agent SDKssmolagents
Build lightweight agents that can use tools and execute code to complete tasks.
Official documentation ↗Hugging Face
Agent SDKsStrands Agents
Build and control agent harnesses with model and tool integrations.
Official documentation ↗Strands Agents
Durable agentsCloudflare Agents
Build stateful agents on Cloudflare with persistent state, scheduling, and real-time interactions.
Official documentation ↗Cloudflare
Choosing where your models run? Explore LLM routers and AI gateways →
How should you choose an agent framework?
Start with your team’s language and the work the agent needs to do. Compare tool integration, state persistence, human approvals, observability, deployment, and how easily you can test a complete task.
How do frameworks relate to MCP?
A framework manages your agent’s execution. MCP provides a standard interface to external tools and context. Check each framework’s documentation for its supported MCP transports and authentication flows.
This is a curated selection, not a popularity ranking or an exhaustive census. Language support and capabilities can change; each entry links to its primary source.