CrewAI is an open-source Python framework for building teams of collaborating AI agents, each with a defined role, goal, and set of tools, that work together to complete complex tasks.
CrewAI is an open-source Python framework for building multi-agent AI systems — teams of AI agents that collaborate on complex tasks rather than a single agent doing everything. Instead of writing one large monolithic prompt, you define multiple agents with specific roles and expertise, assign them tasks, and let them work together as a crew. CrewAI powers over 450 million agentic workflow executions per month and is used by a significant share of Fortune 500 companies for production AI automation.
CrewAI is built around three core concepts: agents, tasks, and crews. These compose into larger systems through flows.
Agents are the individual AI actors in your system. Each agent has a role (its identity within the crew), a goal (what it is optimising for), a backstory (context that influences how it reasons), a set of tools it can call, and an underlying LLM that drives its thinking. A research crew might have a “Senior Research Analyst” agent whose goal is to find verified, primary-source information — this framing shapes how the LLM inside the agent approaches its task.
Tasks define what each agent is asked to do in a given crew run. A task specifies what needs to be produced, which agent is responsible, what the expected output looks like, and whether the task depends on a previous task’s output. Tasks can be assigned to specific agents or left for a manager agent to delegate.
Crews bring agents and tasks together and define how they run. In a sequential process, tasks execute in order: the output of task one becomes context for task two. In a hierarchical process, a manager agent reads the tasks and delegates them to worker agents based on their roles, which more closely mirrors how a real team operates. CrewAI handles the communication between agents, context passing, and result aggregation automatically.
Flows provide event-driven orchestration for more complex systems. A flow can start a crew, wait for its output, branch based on a condition, call regular Python code, and kick off a second crew — all in a readable, decorated Python function. Flows are used when you need state management, conditional logic, or pipelines that span multiple crew runs.
Memory allows agents to retain information across tasks and runs. CrewAI provides short-term memory (shared context within a crew run), long-term memory (a persistent store agents can query across separate runs), entity memory (tracking facts about named people, companies, or objects), and contextual memory (summarising recent interactions). Enabling memory makes agents more consistent and reduces repeated work across related tasks.
Content research and writing pipeline. Define a researcher agent that searches the web and gathers verified facts, a writer agent that drafts a structured article, and an editor agent that refines tone and checks accuracy. Run them as a sequential crew and receive a finished draft in minutes without touching each step manually.
Automated competitive intelligence. A crew of agents that monitors competitor websites, extracts pricing and feature updates, compares them against your own product data, and produces a weekly briefing. Scheduled via a flow, this runs unattended and drops a report to Slack or email automatically.
Customer support triage system. A crew that receives an inbound support ticket, classifies its category, retrieves relevant knowledge base articles, drafts a resolution, and either sends it directly or routes it to a human agent for review when confidence is below a threshold. Human-in-the-loop approval is built into CrewAI’s task callback system.
Automated code review crew. An agent that reads a pull request diff, a second that checks it against your team’s coding standards, a third that suggests specific fixes, and a manager agent that consolidates the feedback into a structured review comment. Useful as a first-pass review before human engineers look at changes.
Sales outreach personalisation. A crew that takes a list of prospects, looks up each company’s recent news and activity, identifies a relevant angle for outreach, and drafts a personalised first message. What would take a sales development team hours to do manually runs in minutes with a three-agent crew.
The CrewAI open-source framework is free under the MIT license with no usage caps or restrictions — you run it in your own environment and pay only for the LLM API calls your agents make. The Enterprise platform (AMP) is commercial: a free tier allows 50 executions per month, a Professional plan costs $25/month for 100 executions, and Enterprise pricing is custom, typically in the range of $60,000–$120,000 per year for larger organisations that need compliance certifications, SSO, and dedicated support. Most developers and small teams build entirely on the open-source version.
Install the CLI with pip install crewai or uv tool install crewai. Run crewai create to scaffold a new project, which generates the folder structure, example agents and tasks, and a .env file for your LLM API keys. Add your OPENAI_API_KEY (or the key for your preferred model), run crewai install to set up dependencies, and then crewai run to execute your first crew. The official documentation at docs.crewai.com walks through agents, tasks, and crews step by step.
For a structured path through building production multi-agent systems — CrewAI, LangChain, and full-stack deployment included — the BuildrLabs Agentic AI Bootcamp covers the complete stack in person in Colombo over 16 weeks.
The CrewAI open-source framework is completely free under the MIT license. You install it with pip, write your agents and crews in Python, and run them against any LLM API — you pay only for the model calls. The Enterprise platform (also called AMP) adds a visual editor, managed deployment, RBAC, and real-time tracing via a commercial subscription. AMP pricing starts with a free tier of 50 executions per month, a Professional tier at $25/month for 100 executions, and custom Enterprise pricing for organisations with compliance and scale requirements.
An agent is a single AI actor with a role (who it is), a goal (what it is trying to accomplish), a backstory (context that shapes how it reasons), and a set of tools it can call. A crew is a group of agents working together on a set of tasks — you define how they collaborate, either sequentially (one after another) or hierarchically (a manager agent delegates to worker agents). A flow is CrewAI's event-driven orchestration layer, used when you need branching logic, conditional paths, or long-running pipelines that combine crews with regular Python code. Most production systems use all three — agents inside crews, crews inside flows.
CrewAI works with any LLM that has a compatible interface, including OpenAI (GPT-4o, o3), Anthropic (Claude), Google (Gemini), Mistral, and locally hosted models via Ollama. You configure the LLM at the agent level, so different agents in the same crew can use different models — for example, a fast, cheap model for a high-volume classification agent and a more capable model for a reasoning agent that synthesises final output.
Supabase is an open-source backend-as-a-service built on PostgreSQL that gives developers a managed database, auth, storage, edge functions, and real-time capabilities under one subscription.
Backend / Database
Vector Agents provides role-based AI digital workers that automate sales, customer support, finance, and HR workflows for business teams.
Assistant
Relay.app is a workflow automation platform that combines a visual builder, built-in AI steps, and human-in-the-loop approval checkpoints for business teams.
Workflow AutomationUpdates from the AI world — what shipped, what we’re using in production, and what’s worth your attention. Two emails a month, no spam.
