CrewAI

CrewAI is an open-source software framework and platform for building AI agents and multi-agent systems, written primarily in Python^[4]. Instead of repeatedly prompting a single model, it lets a developer organise several model-driven agents into a group that works through a task together.

Origins

The project's repository on GitHub was created on 27 October 2023 and is released under the MIT license; it has roughly 60,000 stars^[1]. Its creator is João Moura, who before founding the company was directing AI engineering efforts at Clearbit^[2]. The framework was first released on the Python Package Index in December 2023^[4], and the company behind it launched officially in January 2024^[2].

How it works

CrewAI describes two main building blocks. Crews are groups of agents that collaborate autonomously: each agent can be assigned a role, a goal and tasks, and the group works through them^[4]. Flows are for controlled, event-driven workflows, where the developer fixes the order in which steps run rather than leaving the sequencing to the agents^[4]. The two are meant to be combined — a flow can hand a step over to a crew where open-ended collaboration helps, while the rest of the process stays predictable.

Adoption and funding

TechCrunch reported that the company had attracted about 150 enterprise customers in its first year, and that it pitched itself as an alternative to robotic process automation^[2]. In October 2024 CrewAI announced $18 million in total funding, consisting of an inception round led by boldstart ventures and a Series A led by Insight Partners, with Craft Ventures, Earl Grey Capital and the angels Andrew Ng and Dharmesh Shah also investing^[3]. At the same announcement the company said its open-source offering executed more than 10 million agents per month and was already used by nearly half of the Fortune 500^[3]. TechCrunch reported the company was then valued at around $100 million^[2].

Where it sits

CrewAI does not train its own models; it taps models from vendors such as OpenAI and Anthropic^[2]. That puts it in the layer above the model providers: the model supplies language understanding and generation, while the framework supplies the structure around it — who does what, in what order, and with which tools and hand-offs. This is why the same framework can sit on top of whichever model a team prefers.

The trade-off is that its behaviour depends heavily on the models it calls, and multi-agent designs add their own costs: more calls, more latency, and more places for a run to go wrong. The split between autonomous crews and developer-controlled flows is one way the project tries to manage that, letting teams keep the parts of a process that must be reliable under explicit control.