CrewAI Buyer's Checklist: 12 Checks Before AMP Enterprise

Before you adopt CrewAI or ask for the Enterprise plan of CrewAI AMP (Agent Management Platform), settle twelve checks. The deciding number is 50. The free Basic plan includes 50 workflow executions a month, that is also its maximum, and additional executions are not included, alongside 2 automations on CrewAI cloud only ^[1]. By our arithmetic, that is about 1.7 executions a day on a 30-day month. Enterprise is priced as Custom behind a Request trial button, and it holds the governance layer: SSO, role-based access control, PII redaction, workload identity and customer infrastructure. The verdict is plain. Prototype on the open-source framework or on Basic, and open the Enterprise conversation once identity, data location or volume enters the brief.

Settle what you are actually buying

CrewAI is two products under one name. The open-source CrewAI framework is a Python package under the MIT license: agents, tasks, crews and Flows that you install and run on your own machines without paying CrewAI. AMP is the paid platform that builds crews in Studio, deploys them, triggers them on schedules or app events, and monitors and governs them. Every deployed crew or flow becomes an automation with its own API, triggers, traces and access rules. The first three checks sort out which product you need and who carries which bill.

1. Framework or platform. Running crews and flows locally needs no AMP account at all. A free AMP account covers built-in tracing, and an AMP account is needed for deploying with crewai deploy and for the Tool Repository. AMP app integrations ask for an AMP account with an active subscription, as each integration page says. Most teams end up with an account even when the framework does the work.
2. Where the workload runs. Basic runs on CrewAI cloud only. Enterprise runs on CrewAI or customer infrastructure, with a dedicated VPC and NAT. CrewAI Factory, the self-hosted form of AMP Enterprise, is a containerized deployment for your data center or VPC, on AWS Fargate, Azure Container Apps, Kubernetes via Helm or Red Hat OpenShift. A hybrid of cloud and Factory is also documented.
3. Who pays the model and tool bills. Before deploying you bring your own LLM API keys and any tool keys such as Serper, and those providers bill you directly. Studio also needs an LLM connection with your own OpenAI or Azure key, so even the no-code path runs on your model account.

Open source is not zero cost. By default, memory and knowledge embeddings call OpenAI unless you set another embedder, and many tools, including Tavily, Exa, Firecrawl, Bright Data, E2B and Daytona, need paid keys of their own. A fully local path exists, with Ollama serving both the model and the memory embedder and no external API involved. That suits teams whose data must stay inside. The plan-by-plan cost picture, including the renewal rules, sits in our CrewAI pricing guide.

Test whether Basic can carry your pilot

The fourth check is volume. Basic is a real working plan, not a demo: Studio, tracing, OpenTelemetry export, guardrails, human-in-the-loop input, cron scheduling and export as an MCP server are all on it. What it lacks is room. CrewAI prints no billing period, because it prints no paid price, and the grid below holds the rows from its pricing table that decide most pilots.

4. Two automations and 50 executions. List the automations you need live at the same time and how often each one is scheduled to run. A pilot that needs three automations live at once is already past Basic. During the pilot, watch the usage dashboard, which both plans include, to see how quickly the 50 are spent, and bring that reading to the Enterprise call.

The Enterprise route goes through the Request trial button, and the quote, the execution sizing and the overage terms all come out of that conversation. Bring your own numbers. A usage reading from Basic, a list of the automations you plan to run and the infrastructure you need give the sales team something concrete to size against, and they give you a way to compare the quote with the self-run alternative.

Map the controls your security team will ask for

5. Which controls need Enterprise. The Enterprise plan card adds SSO, RBAC, workload identity, PII redaction and policies on top of everything in the free plan. SSO with Microsoft Entra or Okta, role-based access control, a dedicated VPC and PII redaction for traces are Enterprise only. Redaction also needs deployments on v1.8.0 or higher. RBAC grants combine a principal, a role and a scope, with no deny rules and no custom roles, and automations have no view-only role. Just-in-time SSO maps groups to teams, but it is not SCIM. Group changes apply at the next sign-in.
6. Which certifications cover your setup. In a May 2025 post, CrewAI says CrewAI Factory has a SOC 2 Type II report and a HIPAA Security Rule gap assessment for its secure operations ^[2]. That statement is scoped to Factory. If you plan to run on CrewAI cloud instead, ask which report covers that deployment.

Cost Limit policies are not plan-gated: they can be set on any plan where the Agent Control Plane is enabled, and they sound like a spending brake. They are not. They only email recipients and never pause, throttle or stop a run. The Control Plane shows fleet health, errors, executions, tokens and cost per provider, but only for automations on crewAI v1.13 or newer. The Privacy Policy lists encryption in transit and at rest, role-based access controls and continuous monitoring as safeguards.

Find out what leaves your machines

7. Telemetry in the open-source package. It sends anonymous usage telemetry by default, covering versions, crew and agent metadata, tool usage counts, LLM attributes, task lifecycle and CLI attempts, with exception class names but never error messages . Set CREWAI_DISABLE_TELEMETRY, OTEL_SDK_DISABLED or CREWAI_DISABLE_TRACKING to true, 1, yes or on to turn it off; unrecognized values leave it on.
8. The customer-data clause. Under the Terms you keep ownership of your Customer Data, but you license CrewAI to use it, including AI outputs, to improve its products and services, including training models, and to create aggregated anonymized data . Data pulled from connected third-party services counts as Customer Data. The Privacy Policy separately says CrewAI never uses Google user data for advertising or for training AI models.

Telemetry collects tool names and agent roles, so keep personal information out of them. The share_crew option is an opt-in that sends goals, backstories, task descriptions and outputs, which may hold personal information, to CrewAI to improve the library and train models. Tracing is a different switch: it needs a free AMP account and crewai login, and on the first run a prompt asks whether to share the trace, with nothing uploaded if you decline. Disabling tracing does not stop telemetry, and OTEL_SDK_DISABLED does not stop tracing.

Match the build path to the people who will own it

The framework is code. It needs Python 3.10 or newer and below 3.14, installs with uv tool install crewai, and new projects are JSON-first, with agents in agents/*.jsonc and crew settings in crew.jsonc; the older YAML layout needs --classic. Code execution runs through E2B or Daytona sandbox tools, since CodeInterpreterTool was removed in April 2026. Engineering-led teams will find the deeper setup in CrewAI for software teams.

9. Python skills or Studio. Studio, on both plans, lets people describe a crew in plain language to a Crew Assistant and edit it on a visual canvas. Its export is one-way. Downloaded code cannot be imported back, and redeploying customized code creates a separate automation without the Studio editor, versioning or validation.
10. Your connectors on your plan. Each integration page requires an AMP account with an active subscription. The pricing table puts standard tools and triggers on Basic and enterprise connectors on Enterprise only, without naming which apps fall where, so ask about the exact connector you need. Salesforce needs a one-time admin install of CrewAI's managed package, and the Notion integration only lists users and adds comments.

Studio suits operations people who can describe a process but will not maintain Python, while the framework suits teams that want version control, tests and their own hosting. Mixing them is possible, but the one-way export means choosing an owner early. How to design the agents themselves, from roles and goals to guardrails and human review, is covered in our CrewAI AI playbook.

Independent reviews point the same way. Reviewers on G2 mostly praise how quickly a role-based crew comes together, and their recurring complaints are token use and run time that grow with multi-agent workflows and agents that loop on tools. Budget a hardening phase before a crew carries real work, whichever build path you pick.

Read the support and contract terms before signing

11. Support by plan. Basic gets community support only, through the forum at community.crewai.com; the pricing table lists a support contact only on Enterprise. The Enterprise support rows list dedicated support, Slack or Teams support, a support contact, on-site help, training, deployment, onboarding and development help. A 45-day onboarding is included, and the plan card adds forward-deployed engineering and training a la carte.
12. Refunds, renewal and availability. The Terms say payments are non-cancelable and fees paid are non-refundable. Agreements renew automatically for equal terms unless either side gives written notice at least 30 days before the term ends, and renewal prices can rise with at least 30 days' notice. The Terms promise reasonable efforts on availability and print no uptime percentage.

Two smaller points belong in the same file. During a free Pilot Period either party can end it on written notice, and beta, pilot and trial services can be discontinued at any time and may never become generally available, so a pilot built on a beta feature should have a fallback in the plan. Put the renewal date in a calendar the day you sign. Thirty days pass quickly.

When the Enterprise trial request is worth sending

Send the Request trial form when at least one check lands on the Enterprise side: you need SSO or RBAC, PII redaction on traces, workloads on your own infrastructure (including self-hosted CrewAI Factory) or in a dedicated VPC, enterprise connectors, more than 2 automations or more than 50 workflow executions a month, or dedicated support. If none applies, the open-source framework on your own servers or the free Basic plan covers the pilot, and the remaining bills are your model provider, the tool APIs you choose and, for the self-run path, your own hosting. Settle the telemetry switch and the customer-data clause either way.