CrewAI pricing, free plan and limits

Last updated: 2026-10-13

CrewAI is an MIT-licensed Python framework, started in October 2023, for building multi-agent systems from role-based Crews and event-driven Flows. It supports MCP servers and agent-to-agent (A2A) collaboration, and its CrewAI AMP platform adds a visual editor, tracing, guardrails and governance for production use.

About CrewAI

CrewAI is an open-source Python framework for building teams of AI agents, plus a commercial platform called CrewAI AMP for building, running and governing them. Founder and CEO João Moura, who previously directed AI engineering at Clearbit, created the GitHub repository in October 2023, and the company launched in January 2024. In October 2024 it announced $18M in total funding, with boldstart ventures leading an inception round and Insight Partners leading the Series A. The framework is released under the MIT licence, and TechCrunch described the company as based in San Francisco and Brazil.

The framework has two building blocks. A Crew is a group of agents, each described by a role, goal and backstory, that work through tasks together. A Flow is an event-driven workflow that holds state and can run crews as steps, which gives a team explicit control around otherwise autonomous agents. Agents attach tools, knowledge sources, memory and planning, and can use MCP servers as tools or collaborate over the A2A protocol. On the model side, the docs describe native integrations for OpenAI, Anthropic, Google, Microsoft Azure and Amazon Web Services Bedrock, with LiteLLM as a fallback for everything else, and they name GPT-4o mini as the default model until you set another.

CrewAI AMP is the managed side. It adds Studio, a visual editor with an AI copilot, plus tracing, guardrails, human-in-the-loop input and a usage dashboard, and the Enterprise plan adds single sign-on, role-based access control and deployment in a customer's own VPC. In the AI orchestration category, the closest alternatives are LangChain for a mature, widely adopted framework with the LangSmith observability product, Mastra for TypeScript teams, and n8n for workflow automation where AI is one step among many. Business users who want an assistant without code are better served by Lindy. For a wider buying frame, read Best AI Orchestration Tools in 2026 and Best AI Agent Builders in 2026. Because CrewAI ships its own tracing, the guide on the three jobs inside agent observability helps decide whether that is enough, and the agents and automation group lists the neighbouring categories.

Security and privacy need a deliberate look before a production rollout. Reviewers should request the SOC 2 attestation the vendor publishes, and should know the framework sends anonymous usage telemetry unless it is switched off. Peers with a published SOC 2 claim sit in the SOC 2 facet, and other open-source tools are grouped in the open-source facet.

Pricing

The open-source framework is MIT-licensed and free to self-host, and LLM usage is a separate bill from whichever provider you choose. On the hosted AMP platform, the entry plan costs nothing and the top plan is quoted per customer, so ask for written terms on included executions and overage. By contrast, LangChain publishes per-seat plans for its LangSmith product, while n8n lists cloud tiers on its site.

See the free-tier facet for other tools with a free route.

Key Features

• Crews: Teams of agents defined by role, goal and backstory in YAML, code or both, with optional memory, reasoning, tools and sandbox environments.
• Flows: Event-driven workflows that hold state and run crews as steps, so execution order stays under the developer's control.
• Tools and MCP: Hundreds of open-source tools out of the box, MCP servers as tools over stdio, SSE or streamable HTTP, and sandbox tools for E2B and Daytona.
• Model connections: Native SDK integrations for OpenAI, Anthropic, Google Gemini, Azure and AWS Bedrock, with LiteLLM as the fallback for other providers.
• Memory and knowledge: A memory system the vendor describes as remembering, resolving contradictions and forgetting on purpose, plus knowledge sources agents can query.
• Checkpointing: Captures runtime state at every step so a run can be replayed from a chosen step, forked, or rerun with different inputs.
• Agent-to-agent protocol: Connects crews over the A2A protocol, with native async support and streaming results from flows and crews.
• Studio and tracing: A visual editor with an AI copilot, tracing, OpenTelemetry, guardrails, human-in-the-loop input and a usage dashboard on the cloud platform.

Pros

• The framework is MIT-licensed, and its GitHub repository shows about 59.6K stars and 8.7K forks (read 2026-10-13), which lowers lock-in risk for a self-hosted build.
• The trust center lists a SOC 2 Type 2 report dated June 2026 and a HIPAA audit report from February 2026, which gives security reviewers documents to request.
• Switching telemetry off takes one environment variable, and the docs list every field collected by default.
• G2 lists a 4.3 out of 5 average from 26 reviews (read 2026-10-13), although the profile is unclaimed and the sample is small.

Cons

• Anonymous telemetry is on by default and collects tool names and agent roles, and when share_crew is enabled it also collects task descriptions and backstories that may contain personal data.
• The docs name gpt-4o-mini as the default model, and the OpenAI SDK is a core dependency, so an OpenAI key is the path of least resistance unless you configure another provider.
• The current release needs Python 3.10 to 3.13, and the docs carry a migration section and a guide to running without LiteLLM, so pin versions and read the changelog before upgrading.
• Enterprise pricing is not published, and the pricing page describes a 45-day onboarding with forward deployed engineering and training sold a la carte.

Visit CrewAI Official Website