The best AI agent frameworks for building your own agents (2026)

Open-source frameworks and SDKs developers use to build their own AI agents. Compared on language, license, cost and paid hosting options.

AI agent framework: quick answer

From the 8 AI agent frameworks on Klyvoo, with prices checked on each official site.

Quick comparison

Agent Best for Starting price Pricing model Free trial Standout
LangGraph Stateful long-running agents with fine control Free (open source) Free; pay for model API usage and optional LangSmith hosting Free LangSmith tier Durable execution that recovers from failures
CrewAI Role-based multi-agent teams and workflows Free (open source) Free; pay for model API usage and optional CrewAI AMP plan Free AMP Basic plan Simple role and task model for multi-agent teams
OpenAI Agents SDK Lightweight multi-agent apps with handoffs Free (open source) Free; pay for model API usage Small set of building blocks, easy to learn
Claude Agent SDK Agents that read files, run commands and edit code Free SDK; pay for Claude API use Free SDK; pay for Claude API usage Built-in tools for files, shell and web search
Microsoft Agent Framework .NET and Python teams building production agents Free (open source) Free; pay for model API usage and any Azure hosting Successor to AutoGen and Semantic Kernel, with migration guides
Google Agent Development Kit (ADK) Multi-language agents with easy Google Cloud deployment Free (open source) Free; pay for model API usage and any cloud hosting Five language SDKs with a consistent API
LlamaIndex Document-heavy agents and RAG in Python Free (open source); LlamaParse from $50/mo Free framework; optional credit-based LlamaParse cloud plans Free LlamaParse plan Deep tooling for document ingestion, retrieval and RAG
Mastra TypeScript developers building agents into web apps Free (open source); Cloud Teams $250/mo Free framework; usage-based cloud plans for hosting and observability Free Mastra Cloud Starter Native TypeScript with Next.js, React, Express and Hono integrations

Prices in USD from each vendor’s official pricing page, checked October 2026. Some links may be affiliate links; that never changes our picks.

Showing 8 AI agent frameworks

1. LangGraph

AI agent framework

Read full review

Best for: stateful long-running agents with fine control

Low-level orchestration framework from LangChain for building stateful agents as graphs with durable execution.

StandoutDurable execution that recovers from failures
Watch out forLow-level API takes time to learn
Starting price
Free (open source)
Free trial
Free LangSmith tier
Languages
Python, JavaScript/TypeScript

2. CrewAI

AI agent framework

Read full review

Best for: role-based multi-agent teams and workflows

Python framework for running teams of role-based AI agents as crews, with event-driven Flows for workflow control.

StandoutSimple role and task model for multi-agent teams
Watch out forPython only (3.10 to 3.13)
Starting price
Free (open source)
Free trial
Free AMP Basic plan
Languages
Python

3. OpenAI Agents SDK

AI agent framework

Read full review

Best for: lightweight multi-agent apps with handoffs

Lightweight open-source framework from OpenAI for multi-agent workflows with handoffs, guardrails and built-in tracing.

StandoutSmall set of building blocks, easy to learn
Watch out forDefaults and best support are for OpenAI models
Starting price
Free (open source)
Languages
Python, JavaScript/TypeScript

4. Claude Agent SDK

AI agent framework

Read full review

Best for: agents that read files, run commands and edit code

Anthropic's SDK that packages Claude Code's tools, agent loop and context management as a Python or TypeScript library.

StandoutBuilt-in tools for files, shell and web search
Watch out forWorks with Claude models only
Starting price
Free SDK; pay for Claude API use
Languages
Python, TypeScript

Best for: .NET and Python teams building production agents

Open-source Microsoft framework for agents and multi-agent workflows, and the successor to AutoGen and Semantic Kernel.

StandoutSuccessor to AutoGen and Semantic Kernel, with migration guides
Watch out forGo version is in preview and missing features
Starting price
Free (open source)
Languages
C# (.NET), Python, Go (preview)

Best for: multi-language agents with easy Google Cloud deployment

Google's open-source, code-first toolkit for building, testing and deploying agents in Python, TypeScript, Go, Java and Kotlin.

StandoutFive language SDKs with a consistent API
Watch out forDeployment guides center on Google Cloud
Starting price
Free (open source)
Languages
Python, TypeScript, Go, Java, Kotlin

7. LlamaIndex

AI agent framework

Read full review

Best for: document-heavy agents and RAG in Python

Open-source Python toolkit for building LLM agents and event-driven workflows over your own documents and data.

StandoutDeep tooling for document ingestion, retrieval and RAG
Watch out forTypeScript version was deprecated in April 2026
Starting price
Free (open source); LlamaParse from $50/mo
Free trial
Free LlamaParse plan
Languages
Python

8. Mastra

AI agent framework

Read full review

Best for: TypeScript developers building agents into web apps

Open-source TypeScript framework for AI agents and workflows with memory, tools, evals and a local Studio.

StandoutNative TypeScript with Next.js, React, Express and Hono integrations
Watch out forTypeScript/JavaScript only; needs Node.js 22.18 or later
Starting price
Free (open source); Cloud Teams $250/mo
Free trial
Free Mastra Cloud Starter
Languages
TypeScript, JavaScript

How AI agent framework pricing works

Every AI agent framework on this page is free to download. LangGraph, CrewAI, OpenAI Agents SDK, Microsoft Agent Framework and LlamaIndex use the MIT license. Google ADK uses Apache 2.0 and so does Mastra's core. The Claude Agent SDK is free to install but runs under Anthropic's Commercial Terms.

Your main running cost is the model. You pay your model provider for the API calls your agents make. With the Claude Agent SDK that means per-token Claude API rates billed to your API key. Microsoft Agent Framework and Google ADK only add usage-based cloud pricing if you host on Microsoft Foundry or Google Cloud.

Some frameworks sell optional hosting or tooling. LangSmith Developer is $0 per seat/month with no deployment access and LangSmith Plus is $39 per seat/month with 1 free small serverless deployment. CrewAI AMP Basic is free with 50 workflow executions a month, then Enterprise is custom pricing. Mastra Cloud has a free Starter tier and Teams at $250/month. LlamaParse is free with 10,000 credits and Starter is $50/month. The OpenAI Agents SDK has no official hosting, so you deploy it yourself.

What we check

Languages

We list every supported language and flag versions in preview or deprecated, like Go on Microsoft Agent Framework and TypeScript on LlamaIndex.

License

We check whether the code is under MIT, Apache 2.0 or commercial terms, since that affects how you can use it.

Model choice

We note whether a framework works with many model providers or is tied to one, as the Claude Agent SDK is to Claude models.

True monthly cost

We separate the free framework from paid hosting and model API bills so you can see what you will actually pay.

Production features

We look for memory, tracing, guardrails, human-in-the-loop steps and recovery from failures.

Every fact comes from the vendor’s official site. How we review

How to choose

  • If you need long-running agents with fine control over state: LangGraph. Durable execution, built-in memory and human-in-the-loop steps suit teams building production agents.
  • If you want a team of role-based agents up and running fast: CrewAI. Its simple role and task model sets up multi-agent crews quickly, with Flows when you need more control.
  • If you build web apps in TypeScript: Mastra. It is native TypeScript with Next.js, React, Express and Hono integrations plus built-in memory, workflows and evals.
  • If your agents must read and search large sets of documents: LlamaIndex. It has deep tooling for document ingestion, retrieval and RAG plus 300+ integration packages.

FAQ: AI agent frameworks

What is an AI agent framework?

An AI agent framework is a code library developers use to build their own AI agents. It gives you building blocks such as the agent loop, tools, memory and handoffs between agents. LangGraph models agents as graphs with durable execution, while the OpenAI Agents SDK keeps to a small set of building blocks with tracing and guardrails built in.

Which AI agent frameworks are open source?

Most frameworks on this page are open source. LangGraph, CrewAI, OpenAI Agents SDK, Microsoft Agent Framework and LlamaIndex are MIT licensed. Google ADK is Apache 2.0 and so is Mastra's core, though its enterprise features sit under a separate license. The Claude Agent SDK is the exception: its Python repo is MIT but use is governed by Anthropic's Commercial Terms.

How do I choose an AI agent framework?

Start with your language and your model. TypeScript teams lean toward Mastra, .NET teams toward Microsoft Agent Framework and Google ADK covers the most languages with Python, TypeScript, Go, Java and Kotlin. If you need fine control over long-running agents, LangGraph fits best. If you want multi-agent teams with less setup, CrewAI is easier to start with. LangChain vs LangGraph →

Are AI agent frameworks free?

The frameworks are free, but you pay for the model API calls your agents make. Paid plans cover optional hosting and tooling: LangSmith Plus is $39 per seat/month, Mastra Cloud Teams is $250/month, LlamaParse Starter is $50/month and CrewAI AMP Enterprise is custom pricing. Free hosted tiers exist too, such as CrewAI AMP Basic with 50 workflow executions a month and Mastra Cloud Starter.

Head-to-head comparisons

Side-by-side guides from the Klyvoo team on the tools in this list.

More jobs AI agents can do

AI receptionistsAI scheduling assistantsAI sales agentsAI customer support agentsAI content and social agentsAI research agentsAI coding agentsAI personal assistantsAI lead generation agents

Every price on Klyvoo comes from the vendor’s official pricing page. How we review