CrewAI

Agent frameworks

Open-source Python framework for role-based multi-agent crews, with a hosted build and runtime that adds enterprise governance, native SDKs for the major LLMs, and Ollama via LiteLLM.

Verified 2026-08-23 · theagentsindex.com

From: Free (month (50 workflow executions))
Pricing model: freemium
Free tier: Yes
Deployment: Both
Interface: Library
Open source: Yes
Public API: Yes
Model / LLM: Native SDKs for OpenAI, Anthropi…

Our verdict

CrewAI is the most-used open-source Python framework for role-based multi-agent systems, with the GitHub repo sitting at over 57,000 stars.

The framework itself is MIT-licensed, free, and self-hostable from PyPI or GitHub with no execution limit. The hosted build and runtime product is two tiers: free Basic to 50 workflow executions a month, then custom Enterprise with SSO, RBAC, PII redaction hooks, dedicated VPC and a 45-day onboarding.

There is no self-serve paid tier between them, which is the main friction for teams that want a paid plan with a published price.

Production case studies from Docusign, Piracanjuba, Gelato and a leading food ordering service describe measurable gains; reviewers on a Reddit thread note that for tinkering and small projects the framework is fun but often overkill, and that LangGraph is better supported for finer state control.

Teams comfortable in Python who want a framework with a clear role-based abstraction, a stable GitHub presence, native SDKs for OpenAI, Anthropic, Gemini, Azure and Bedrock, and an Ollama path for local models, get real flexibility. Teams wanting a transparent self-serve paid tier or explicit state-machine control over every transition will need to look elsewhere.

What is CrewAI?

CrewAI is an MIT-licensed Python framework for orchestrating role-based multi-agent systems, paired with a hosted build and runtime that adds a no-code visual editor, governance and a control plane. The framework itself is free and self-hostable from GitHub, with Sequential and Hierarchical as the two coordination modes between agents.

The hosted product ships as two published tiers: a free Basic plan capped at 50 workflow executions a month, and a custom Enterprise plan. Native SDK integrations cover OpenAI, Anthropic, Google Gemini, Azure, AWS Bedrock and Snowflake Cortex. Every other supported model, including Hugging Face, Mistral, Groq, Perplexity, Cerebras, SambaNova and local models through Ollama, routes through LiteLLM. CrewAI reports 450 million agentic workflows a month, 4,000 sign-ups a week and use by 65 percent of the Fortune 500.

What does CrewAI do?

CrewAI lets developers define teams of AI agents, called crews, that collaborate using one of two coordination modes: Sequential, where agents execute one after another, or Hierarchical, where a Crew Manager agent delegates work to team members.

Building custom agents requires Python knowledge, plus familiarity with scripting, environment variables and dependency management; agents retain memory and context across tasks within a workflow, though they do not learn or improve from past executions.

The hosted product adds a no-code visual editor that exports to Python, an AI copilot, GitHub integration for version control, and roughly twenty premade templates covering content management, prospect analysis and lead scoring. A code-first API lets teams build custom integrations, and both the Free and paid plans let workflows be exported as an MCP server or a UI component.

On the Enterprise tier, the Control Plane adds RBAC, immutable audit trails, SSO, workload identity, and runtime hooks that inject PII redaction and policy checks at every LLM and tool call, with human-in-the-loop approval gates during execution. Enterprise deployments run on CrewAI cloud, a customer's own VPC, or fully on-premises infrastructure, with a 45-day onboarding program.

How CrewAI works

Define agents and tasks in Python, or use the no-code visual editor, and choose Sequential or Hierarchical collaboration.
Connect LLMs and tools, then export the crew as an MCP server or a UI component.
Deploy via the hosted platform or self-host the open-source framework, tracking executions against your plan's quota.

Key features

Sequential and Hierarchical crews: Agents run one after another, or a Crew Manager agent delegates tasks to team members in a hierarchical structure.
No-code visual editor: Build workflows visually on the hosted product, then export the result to Python code.
Control Plane governance: Enterprise tier adds RBAC, immutable audit trails, Enterprise IAM, SSO and workload identity for compliance.
Runtime policy hooks: Hooks inject PII redaction and policy checks at every LLM and tool call during execution.
MCP server export: Workflows on both Free and paid plans can be exported as an MCP server or a UI component.
Native SDK integrations: Native provider SDKs for OpenAI, Anthropic, Google Gemini, Azure, AWS Bedrock and Snowflake Cortex, with LiteLLM routing every other supported model including Hugging Face, Mistral, Groq, Perplexity, Cerebras, SambaNova and Ollama for local models.
Premade templates: Roughly twenty templates are available for content management, prospect analysis and lead scoring.

What are CrewAI's use cases?

Lead enrichment at scale: Gelato used CrewAI agents to enrich over 3,000 leads per month.
Faster lead response: Docusign achieved 75 percent faster first contact with leads using CrewAI agents.
Customer support accuracy: Piracanjuba reached 95 percent response accuracy for customer support using CrewAI agents.
Voice agent QA automation: A food ordering service cut voice agent QA time from 74 hours to 3 hours, a 96 percent reduction, using CrewAI.
Faster internal development: General Assembly reported a 90 percent reduction in development time building with CrewAI agents.

Who is CrewAI for?

Python developers building custom multi-agent workflows
Enterprises needing governed, auditable agent deployments with SSO and RBAC
Teams wanting a no-code starting point that still exports to Python

Not forTeams wanting a transparently-priced self-serve paid tier: CrewAI jumps directly from a 50-execution free Basic plan to a custom Enterprise sales conversation, with nothing published in between.

What does CrewAI integrate with?

OpenAI (native SDK)
Anthropic (native SDK)
Google Gemini API and Vertex AI (native SDK)
Azure AI Inference and Azure OpenAI (native SDK)
AWS Bedrock (native SDK, Converse API)
Snowflake Cortex (native SDK)
Hugging Face, Mistral, Groq, Perplexity, Fireworks AI, SambaNova, Cerebras, Open Router, Nebius AI Studio, IBM watsonx.ai, Meta Llama API, NVIDIA NIM and SageMaker, all routed through LiteLLM
Ollama for local models, routed through LiteLLM
GitHub integration for sign-up and version control
MCP server export and MCP client adapter

Why use CrewAI?

Docusign reported 75 percent faster first contact with leads using CrewAI agents
Piracanjuba reached 95 percent response accuracy for customer support with CrewAI agents
A food ordering service cut voice agent QA time by 96 percent, from 74 hours to 3 hours
General Assembly reported a 90 percent reduction in development time using CrewAI agents

What are CrewAI's pros and cons?

What's great

MIT-licensed open-source framework, free to self-host with unlimited crews and executions, currently 57,000+ stars on GitHub
Native SDK integrations for OpenAI, Anthropic, Google Gemini, Azure, AWS Bedrock and Snowflake Cortex, with LiteLLM routing every other supported provider plus Ollama for local models
Control Plane governance on Enterprise adds RBAC, immutable audit trails, SSO, workload identity, PII redaction hooks and human-in-the-loop approval gates
Customer case studies report measurable gains: 75 percent faster lead contact at Docusign, 95 percent response accuracy at Piracanjuba, 96 percent reduction in QA time for a voice agent deployment, 90 percent reduction in development time at General Assembly
Flexible deployment: hosted CrewAI cloud, customer's own VPC, or fully self-hosted open source

Watch-outs

Free Basic is capped at 50 workflow executions a month and the only path beyond that is a custom Enterprise sales conversation. There is no published self-serve paid tier.
Building custom agents requires Python, scripting, environment variables and dependency management, despite the no-code visual editor
Agent collaboration is limited to two coordination modes, Sequential and Hierarchical, with no built-in graph-style state control for finer orchestration
Agents retain memory and context within a workflow but do not learn or improve from past executions
Pricing jumps directly from a free 50-execution Basic tier to custom Enterprise, with no self-serve plan in between
Tight, deterministic control over long-running orchestration needs Flows rather than the crew abstraction alone

CrewAI pricing

Free Basic to 50 workflow executions a month. Enterprise is custom-priced. CrewAI publishes no self-serve tier between them.

Basic
Free (month (50 workflow executions))
- Visual editor and AI copilot
- GitHub integration
- 50 workflow executions/month
- MCP server and UI component export
Enterprise
Custom pricing
- SSO (MS Entra, Okta), RBAC, workload identity
- PII redaction and policy hooks at every LLM and tool call
- Dedicated VPC, FedRAMP High, on-prem deployment
- Human-in-the-loop approval gates, hallucination scores, deployment history
- Deploy on CrewAI cloud, your own VPC, or your own infrastructure
- 45-day onboarding, forward-deployed engineering and training available a la carte

See current pricing on crewai.com ↗Compare CrewAI alternatives →

Frequently asked questions

Is CrewAI free to use?

The open-source framework can be self-hosted for free with unlimited crews and executions. The hosted product also has a Free Basic plan limited to 50 workflow executions per month.

Do I need to know how to code to use CrewAI?

The hosted product offers a no-code visual editor that exports to Python, but building custom agents still requires Python knowledge, scripting, environment variables and dependency management.

What LLMs does CrewAI support?

CrewAI has native SDK integrations for OpenAI, Anthropic, Google Gemini, Azure, AWS Bedrock and Snowflake Cortex. Every other supported provider, including Hugging Face, Mistral, Groq, Perplexity, Cerebras and local models through Ollama, is routed through LiteLLM.

Does CrewAI offer an API?

Yes. CrewAI provides a code-first API for total control, letting teams build custom integrations that connect crews with other applications.

What happens if I exceed my plan's execution quota?

The Basic plan is capped at 50 workflow executions a month. Beyond that, CrewAI does not offer pay-as-you-go overage billing, so the path is to upgrade to a custom Enterprise contract.

CrewAI alternatives

LangChain
The larger, more mature orchestration framework, with LangGraph's native graph-branching, transparent LangSmith observability pricing and the widest integration ecosystem, for teams building a bespoke production agent that needs finer state control than CrewAI's role-based crew abstraction.
Pydantic AI
A minimal, type-safe framework centered on catching malformed model output at write-time, for teams whose problem is validated structured output rather than standing up a team of collaborating agents, though CrewAI's managed cloud runtime (tracing, guardrails, RBAC) still needs less custom infrastructure to get running.
Agno
A faster-positioned, similarly free-and-open framework with memory, knowledge and MCP built in, but its paid production layer is BYOC-only rather than CrewAI's fully managed cloud runtime.
LangGraph
A lower-level graph engine with checkpointed state, replay and human-in-the-loop interrupts, for teams that have outgrown CrewAI's role-based abstraction and need explicit control over every transition.
Compare the whole category
Every agent in Agent frameworks, side by side on the same fields.

Visit CrewAIOpens the vendor's own site Compare CrewAISide by side with the category See alternativesOther options worth weighing

Anything to add?

This page is researched and re-verified by our editors — but the people who actually use CrewAI know things we cannot see from the outside.

Write a reviewUsed CrewAI? Your experience helps the next buyer decide.

Get a quote from CrewAITell us what you need and we make the introduction.

Get weekly Agent frameworks updates

New agents and fresh verdicts. One short email, every week. One-click unsubscribe.

Advertise

Reach buyers mid-decision. Put your brand in front of them — sponsor slots are open to any advertiser (see the content policy); listed agents can go Featured at the top of their category.

Featured — top of your category →Sponsor this page →

The evidence

Facts re-verified 4 days ago

How we checked this

CrewAI's own pricing page names only two tiers: Basic is free for up to 50 workflow executions a month, and Enterprise is custom-priced. CrewAI's own site does not publish any figure between free and custom.