# How to put a human check into a CrewAI Crew or Flow

CrewAI is MIT-licensed Python with three documented ways to pause for a person, one of them labelled Enterprise. Its tutorial needs an LLM key and a Serper key, and its pricing page shows a free tier of 50 executions a month. Read 2026-11-06.

Short answer, from the sources below: CrewAI is a Python library whose repository carries the MIT licence, for running teams of AI agents ("Crews") inside a fixed sequence of steps ("Flows"). Its documentation describes three ways to stop and ask a person, and the guide's table lists the webhook route, labelled Enterprise, as the one for production deployments. Its beginner tutorial writes a research report and needs two outside keys. The pricing page shows a free tier and an Enterprise tier with the price left to a sales call. This page lists what each source says, who wrote it and what it leaves out.

## What it is

CrewAI's documentation says: "CrewAI is the leading open-source framework for orchestrating autonomous AI agents and building complex workflows."[1] The "leading" is the company's claim. It splits the work in two: "Flows allow you to create structured, event-driven workflows that manage state and control execution", and "Crews are teams of autonomous agents that collaborate to solve specific tasks delegated to them by the Flow."[1]

The repository is marked "MIT license", states "CrewAI requires Python >=3.10 and <3.14", and showed 60.6k stars when read.[2] It also sends usage data by default: "CrewAI uses anonymous telemetry to collect usage data". The README says no data is collected on "prompts, task descriptions, agents' backstories or goals, usage of tools, API calls, responses, any data processed by the agents, or secrets and environment variables, with the exception of the conditions mentioned". One condition is the `share_crew` feature: "When the share_crew feature is enabled, detailed data including task descriptions, agents' backstories or goals, and other specific attributes are collected". It can be turned off: "Users can disable telemetry by setting the environment variable OTEL_SDK_DISABLED to true."[2]

## Three ways to put a person in the loop

CrewAI's guide to human-in-the-loop work lists two approaches in a table, "Flow-based (@human_feedback decorator)" and "Webhook-based (Enterprise)".[3] The docs describe a third, simpler one on a separate page. All three are in the documentation version read (v1.15.34).

1. **A flag on a task.** "Set the human_input flag in the task definition. When enabled, the agent prompts the user for input before delivering its final answer."[4]
2. **The `@human_feedback` decorator in a Flow.** It "requires CrewAI version 1.8.0 or higher" and lets a Flow pause, show output to a person, collect feedback and route on the outcome. A model reads the feedback and picks the route: the `llm` setting is the "LLM used to interpret feedback and map to an outcome."[5] By default it waits at a terminal: "By default, @human_feedback blocks execution waiting for console input." A custom provider can make it non-blocking for "Slack, email, webhooks, or APIs."[5]
3. **Webhooks on the Enterprise platform.** A crew started with a webhook address stops with "the Pending Human Input state"; a person answers by calling the resume endpoint, and "Webhook URLs Must Be Provided Again" in that call, because they are not carried over from the kickoff.[3]

The guide labels only the third route Enterprise; it does not say what the first two cost or require beyond the library.[3][4] The pricing page lists a row "Human-in-the-loop input" marked Yes under both the free and the Enterprise plans, and does not say which of the three it means.[7]

## What the tutorial needs

The quickstart promises "Build your first CrewAI Flow in minutes" and gives no duration. Its prerequisites are a Python environment and the CrewAI command line, "An LLM configured with the right API keys", and "A Serper.dev API key (SERPER_API_KEY) for web search in this tutorial". The steps are `crewai create flow`, `crewai install` and `crewai run`, and the result is a markdown report on disk, `output/report.md`, on a researcher agent's web searches.[6] The tutorial is a research task; it does not touch invoices, a spreadsheet or a client system.[6]

## What the platform says it charges

CrewAI's pricing page shows two plans. The free one, labelled "BASIC", lists a "Visual editor and AI copilot", "GitHub integration" and "50 workflow executions/month". The other, "ENTERPRISE", is "Custom": it adds "SSO, RBAC, workload identity, PII redaction, and policies", a choice to "Deploy on CrewAI cloud, your own VPC, or your own infrastructure", and a "45-day onboarding".[7] Both plans list Tracing and OpenTelemetry; SSO and role-based access appear only under Enterprise.[7] The page shows no price in dollars and no tier between the two.[7] The [monitoring and audit trails page](/research/ai-agent-monitoring-audit-trails) covers what OpenTelemetry records.

## Not yet proven

- **That the tutorial runs.** Nothing was installed or run, and the docs give no duration for it.
- **What a run costs in model and search fees.** The tutorial needs an LLM key and a Serper key; the pages state no price for either.
- **Which human-input route the free plan includes.** The pricing row says Yes and names no route.
- **What the free tier counts.** The page says "50 workflow executions/month" and the definition of an execution was not read.
- **Prices beyond Enterprise.** The pricing page read lists no figure for any plan.
- **Telemetry in practice.** The README describes what is collected; the traffic was not inspected.
- **Comparisons with other frameworks.** LangChain and the OpenAI Agents SDK are covered on their own pages and not compared here.

If you build agents for clients and have put a human check into one, I would like to compare notes: mackenzie@mackenziebowes.com.

## Sources

1. CrewAI documentation (v1.15.34), Introduction. Read in Rotunda 2026-11-06. <https://docs.crewai.com/v1.15.34/en/introduction>
2. crewAIInc/crewAI repository README. Read in Rotunda 2026-11-06. <https://github.com/crewAIInc/crewAI>
3. CrewAI documentation (v1.15.34), Human-in-the-Loop (HITL) Workflows. Read in Rotunda 2026-11-06. <https://docs.crewai.com/v1.15.34/en/learn/human-in-the-loop>
4. CrewAI documentation (v1.15.34), Human Input on Execution. Read in Rotunda 2026-11-06. <https://docs.crewai.com/v1.15.34/en/learn/human-input-on-execution>
5. CrewAI documentation (v1.15.34), Human Feedback in Flows. Read in Rotunda 2026-11-06. <https://docs.crewai.com/v1.15.34/en/learn/human-feedback-in-flows>
6. CrewAI documentation (v1.15.34), Quickstart. Read in Rotunda 2026-11-06. <https://docs.crewai.com/v1.15.34/en/quickstart>
7. CrewAI, Pricing. Read in Rotunda 2026-11-06. <https://crewai.com/pricing>

---

Published 2026-11-06. Web page: https://www.mackenziebowes.com/research/crewai-human-input-and-pricing
