CrewAI Pricing: Free Basic Plan, Custom Enterprise, Real Costs

CrewAI costs nothing to start, in two separate ways, and CrewAI prints no dollar price anywhere. The open-source CrewAI framework is MIT-licensed, so you can install it and run crews and flows on your own machines without paying CrewAI. The hosted platform, CrewAI AMP (Agent Management Platform), has a free Basic plan with 2 automations and 50 workflow executions a month, plus an Enterprise plan sold at a custom price on request ^[1]. The money you actually spend goes elsewhere. Your LLM provider bills every model call, tool APIs bill their own keys, and self-hosting is your own server bill. Budget for tokens first, CrewAI second.

Two plans, no price tags

AMP has exactly two plans. Basic is listed as Free and opens with a Sign Up button, while Enterprise is listed as Custom and opens with a Request trial button. CrewAI's pricing page prints no dollar price for Enterprise; it is sold on request. Most of the product is shared between the two. Studio, the no-code builder, sits on both plans, as do standard tools and triggers, private agent and tool repositories, workflow templates, GitHub integration, export as an MCP server or UI component, tracing, OpenTelemetry, AI agent training, LLM testing, guardrails, human-in-the-loop input, workflow chat in the UI and in Slack and Teams, the usage dashboard, token count, performance metrics, deployment history, LLM management, automatic scaling and cron scheduling.

The Basic ceiling is firm on paper. Basic includes 50 workflow executions a month, the maximum is also 50, and additional executions are not included on Basic. By our arithmetic, that is about 1.7 executions a day across a 30-day month, enough to prove that a deployed automation works and too little to run a team's daily process on. The two-automation limit bites just as early, because every crew or flow you deploy to AMP becomes its own automation with its own API, triggers and traces.

What only Enterprise buys

Enterprise is where governance lives. The plan card lists SSO, role-based access control, workload identity, PII redaction and policies on top of everything in Basic. Of the policies, only PII redaction is gated: PII redaction for traces needs the Enterprise plan, and creating PII redaction policies in the Agent Control Plane does too, while lower plans see that editor read-only. Cost Limit policies are not plan-gated (see cost alerts below). Deployment options widen to CrewAI cloud, your own VPC or your own infrastructure, including CrewAI Factory, the self-hosted form of AMP Enterprise. A 45-day onboarding is included. The support rows list training, deployment, onboarding and development help on Enterprise, and the plan card adds that forward-deployed engineering and training are also available a la carte.

Where the real money goes: models, tools and servers

Neither plan includes model usage. Before you deploy to AMP, you need your own LLM API keys (OpenAI, Anthropic, Google and others) plus keys for any tool APIs such as Serper, and those providers bill you directly ^[2]. Studio works the same way. It needs an LLM connection on your own OpenAI or Azure key first, so every token it spends lands on your provider account. The LLM management row on the pricing table is a feature for handling those connections, not a bundled allowance. Snowflake customers can route models through their own Snowflake Cortex access instead.

Running the framework yourself has the same shape. Open source is not zero cost. The default model is OpenAI's, so the default path needs an OpenAI key, and memory embeddings and crew planning also call OpenAI unless you set another embedder or a separate planning model. Tools add their own meters: Tavily, Exa, Firecrawl, Bright Data, Browserbase, and the E2B and Daytona sandboxes that now handle code execution all need their own accounts and keys. Serper offers a free tier. Hosting is your server or cloud bill. A fully local path exists, with Ollama for the model and an Ollama embedder for memory; our guide to whether CrewAI is free walks through it.

Multi-agent work is where token bills climb. Reviewers on G2 report that token use and run time grow quickly with multi-agent workflows, and that agents can loop on tools. CrewAI gives you direct spend controls in the framework: max_iter, max_execution_time and max_rpm on agents, and max_tokens on tasks, with a crew-level max_rpm overriding the agents. Token usage is reported per crew, and a flow can total its own usage. The fifth identical tool call, meaning the same tool with the same arguments and the same result, also stops the task, a fixed limit you cannot change. Set those caps before the first long job, not after the first invoice.

Cost alerts on AMP

The Agent Control Plane shows tokens and cost per provider over 24 hours, a week or 30 days, for automations running crewAI v1.13 or newer. Cost Limit policies are not plan-gated: you can create one on any plan where the Agent Control Plane is enabled. They are alerts, not brakes. A policy emails its recipients when an automation's LLM spend passes a daily, weekly or monthly budget, and it never pauses, throttles or stops a run ^[3]. Thresholds apply per automation rather than summed across the scope, and the check runs periodically, so the email arrives with a short delay.

Billing period, renewal and refunds

CrewAI prints no billing period because it prints no paid price. There is no monthly or annual figure for either plan, so the usual monthly-versus-yearly arithmetic has nothing to work on. Paid terms live in a contract instead. CrewAI's Terms say fees are set in an Order Form, payments are non-cancelable and fees paid are non-refundable ^[4]. Amounts exclude taxes, which the customer pays. Read that before signing, not after.

Renewal is automatic. An Order Form agreement renews for a term of equal length unless either side gives written notice of non-renewal at least 30 days before the current term ends. CrewAI can also raise prices for a renewal term, provided it gives at least 30 days' notice before the term ends. Those two clauses decide your real exit window. In practice, the date that matters is not the renewal date itself but the day 30 days before it, because a notice sent later rolls you into another full term at whatever price was announced.

How the bill compares with n8n and LangSmith

With no CrewAI price to set against a rival's, the honest comparison is between billing models. Open-source CrewAI charges nothing for the framework. AMP meters workflow executions and automations by plan, its table has no seat row, and model usage is always billed separately by your own provider. Other agent platforms put the meter in different places, so match the meter to how your team works before you compare anything else.

n8n also bills on workflow executions, counting a full run from start to finish, with no per-step or per-user charge. Its self-hosted Community Edition is free. n8n Cloud Starter costs $20 a month billed annually or $24 billed monthly, for 2,500 executions a month, and n8n AI steps also bill model tokens, through Gateway credits or your own key. n8n's executions and AMP's workflow executions are separate units defined by separate vendors, so the counts do not line up one to one. What carries over is the shape: both meter executions rather than seats, and both leave model tokens on a second bill.

LangGraph, LangChain's graph framework, is also MIT-licensed; its managed deployment runs through LangSmith, which bills by the seat plus pay-as-you-go traces and deployment resources. LangSmith Developer costs $0 per seat a month for 1 seat, and LangSmith Plus costs $39 per seat a month. By our arithmetic, a 3-seat Plus team pays $117 a month for seats alone, before pay-as-you-go usage. LangSmith Cloud deployments need Plus or Enterprise and are billed on the compute and database resources each deployment requests, with no rate printed; Plus includes 1 free small serverless Agent Server. Seat-based billing grows with headcount; execution-based billing grows with workload. Pick the one your growth will hit last.

Self-run, Basic or a quote: matching the plan to the workload

If your team writes Python and already runs servers, the open-source framework costs you nothing beyond tokens, tool keys and hosting, and it needs no AMP plan to run locally. Basic suits the step after that: deploying one or two automations to CrewAI cloud, testing Studio and triggers, and seeing traces, within 50 workflow executions a month, with community support only. Ask for an Enterprise quote once you need more than 2 automations or more than 50 workflow executions a month, enterprise connectors, SSO, role-based access control, PII redaction, a dedicated VPC and NAT, or dedicated support. Before that call, our CrewAI buyer's checklist lists the questions to settle first.