Drydock.si
Submit a tool

// category · 18 tools · 18 open source

AI Agent Frameworks

Libraries for composing models, tools, memory and control flow into agents. Pick on language, abstraction level, observability and how much control you need over the loop.

An AI agent framework is the code library that runs the loop at the heart of every agent: send the conversation to a model, run any tool the model asks for, feed the result back and repeat until the task is done or a limit is hit. The frameworks listed here differ in how much of that loop they hide, how they keep state between steps, and which language you write in.

Python has the widest choice. LangGraph models an agent as a graph of nodes with checkpoints, which suits long tasks that need branches, retries and human approval. CrewAI organizes work as a crew of role based agents. Pydantic AI validates every output against a typed model, smolagents lets the model write Python to call tools, and the OpenAI Agents SDK keeps the primitives small: agents, tools, handoffs and guardrails. DSPy is a different kind of tool: it tunes the prompts inside a pipeline against a metric. Our best AI agent framework for Python guide compares the main options with code.

TypeScript teams usually choose between Mastra, a full framework with workflows, memory and a local studio, and the Vercel AI SDK, a lighter library whose version 7 adds tool approvals and durable agents. See the best TypeScript agent frameworks roundup and our Vercel AI SDK guide.

When you compare frameworks, check four things. First, state: can a run survive a crash or wait hours for an approval? Second, tools: does it speak the Model Context Protocol, so one tool server works everywhere? Third, observability: can you see every model call and tool call in a trace? Fourth, lock in: can you swap model providers without rewriting your agent? Most frameworks listed here are open source and free; you pay for the models they call and for any hosted platform the vendor sells on top, such as hosted deployment and enterprise plans. Most agents built with these frameworks also need a web search tool; Tavily vs Exa vs Firecrawl compares the three APIs we see most.

For the difference between a framework and the runtime around it, read what an agent harness is, and for a head to head on the three most popular Python options, see LangGraph vs CrewAI vs OpenAI Agents SDK. Official references: the LangGraph overview and the AI SDK agents docs.

LlamaIndexllamaindex.aiFramework for building agents over your data, strong on ingestion, indexing, retrieval and document workflows.Agent FrameworksOpen source★ 52.7kAgnoagno.comPython framework and runtime for multiagent systems with memory, knowledge and tools.Agent FrameworksOpen source★ 42.9kLettaletta.comPlatform for stateful agents with long term memory, evolved from the MemGPT research project.Agent FrameworksOpen source★ 25.3kDSPydspy.aiFramework for programming rather than prompting LLMs, with optimisers that tune prompts and weights automatically.Agent FrameworksOpen source★ 38.8kMastramastra.aiTypeScript framework for agents, workflows, RAG and evals, built by the team behind Gatsby.Agent FrameworksOpen source★ 29.1kAgent Development Kitadk.devGoogle's open source framework for developing, evaluating and deploying multiagent systems.Agent FrameworksOpen source★ 22kLangChainlangchain.comWidely used framework for building LLM applications and agents, with hundreds of model, tool and data integrations.Agent FrameworksOpen source★ 148.6kAutoGenmicrosoft.github.ioMicrosoft Research framework for building multiagent systems where agents converse to solve tasks.Agent FrameworksOpen source★ 61.6kCrewAIcrewai.comPython framework for orchestrating role playing teams of agents that collaborate on tasks, plus an enterprise platform.Agent FrameworksOpen source★ 60.2kLangGraphlangchain.comLow level orchestration library for stateful, long running agents modelled as graphs, with persistence and human in the loop.Agent FrameworksOpen source★ 44kOpenAI Agents SDKopenai.github.ioLightweight framework for multiagent workflows with handoffs, guardrails and built in tracing.Agent FrameworksOpen source★ 30.3ksmolagentshuggingface.coHugging Face's minimal library for agents that think in code, working with any model.Agent FrameworksOpen source★ 30.1kSemantic Kernellearn.microsoft.comMicrosoft's model agnostic SDK for integrating LLMs and agents into C#, Python and Java applications.Agent FrameworksOpen source★ 28.7kVercel AI SDKai-sdk.devVercel AI SDK: a free, open source TypeScript toolkit with one API for many model providers, streaming chat hooks and the ToolLoopAgent class.Agent FrameworksOpen source★ 27.6kHaystackhaystack.deepset.aiOpen source orchestration framework from deepset for production grade RAG pipelines and agents.Agent FrameworksOpen source★ 26.9kPydantic AIpydantic.devAgent framework from the Pydantic team that brings type safety and structured outputs to LLM apps.Agent FrameworksOpen source★ 20.9kVoltAgentvoltagent.devOpen source TypeScript agent framework with memory, workflows, tools and MCP support, paired with the optional VoltOps console for tracing and deployment.Agent FrameworksOpen source★ 10.9kDocker Agentdocs.docker.comDocker's open source runtime, formerly cagent, that runs AI agents and multiagent teams defined in YAML or HCL, with MCP tools, sandboxing and registry sharing.Agent FrameworksOpen source★ 4.6k

// guides

Guides to AI agent frameworks

All guides →

More categories

Orchestration & WorkflowsVector DatabasesInference & HostingCoding AgentsDev Tools & Protocols