Self-host CrewAI on a Cube

CrewAI is an open-source Python framework for orchestrating role-playing, autonomous AI agents that collaborate on tasks. It is MIT-licensed with about 58,500 stars on GitHub (verified 2026-09-13) and offers two complementary primitives: Crews (teams of agents) and Flows (event-driven workflows). A Cube is a good place to run it: you get a long-lived Python framework with no per-seat pricing.

The install is one uv command. The part worth reading carefully is Step 4 — CrewAI is a framework, not a finished agent. The first crewai create command builds a starter project that you then customize; the install is the easy half.

Before you start

CrewAI is a Python framework on top of uv, the Astral package manager. It needs Python 3.10 or newer (and below 3.14 at the time of writing). The framework itself is small; the heavy resource is the LLM each agent calls.

• A Cube of 2 vCPU, 4 GB RAM, 40 GB disk — about $0.0140/hour, roughly $0.34/day or $10/month if left running. Comfortable for the framework plus a small local model.
• If you want to run a capable local model (around 7B parameters), plan on 8 GB RAM and 8 GB of disk for the model weights alone, in addition to the agents. The Cube size for that is 4 vCPU, 16 GB RAM, 100 GB disk.
• An SSH key pair, as with any Cube.
• An API key from an AI provider — or a local Ollama install if you would rather skip the paid API.

Step 1 — Create the Cube

Create a Cube with the Ubuntu 24.04 image. CrewAI runs on Linux; Ubuntu 24.04 is the image Krova validates against. uv is installed by the CrewAI installer if it is not present.

Size it at 2 vCPU / 4 GB / 40 GB if you are pointing CrewAI at a paid AI provider, or 4 vCPU / 16 GB / 100 GB if you want to run a local 7B model through Ollama on the same Cube.

Step 2 — Connect over SSH

ssh ubuntu@<cube-host> -p <port>

Step 3 — Install the CrewAI CLI

The CLI installs globally with uv:

uv tool install crewai

If uv is not on the Cube yet, install it first:

curl -fsSL https://astral.sh/uv/install.sh | sh

Confirm the install:

crewai --version

Step 4 — Create a starter project

CrewAI ships a project scaffold. Pick a name and a template:

crewai create crew my-crew

The CLI asks a few setup questions — project name, default model provider, which tools to enable. Use --classic if you want the older Python/YAML scaffold rather than the new uv-based one.

Move into the project and run it:

cd my-crew
crewai install
crewai run

The first run pulls the dependencies uv needs for the project and launches the crew. Watch the logs — each agent in the crew prints what it is doing and what it is handing off.

Step 5 — Pick the model backend

CrewAI's agents call out to any OpenAI-compatible provider. Set the relevant environment variables before launching the crew.

Paid API

Set OPENAI_API_KEY (or the equivalent for Anthropic, Groq, etc.) and crewai run uses the provider's hosted models. Billing is on the provider side.

Local Ollama on the same Cube

For a no-per-message bill, install Ollama on the same Cube and point CrewAI at it. Ollama exposes an OpenAI-compatible API on http://localhost:11434:

curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3.1:8b

Then set OPENAI_API_BASE=http://localhost:11434 and pick the llama3.1:8b model in the crew configuration. Everything from there is local.

Step 6 — Leave it running

CrewAI crews and flows are typically long-lived — they wake on a schedule, listen for a webhook, or wait for an upstream event. Run the project under tmux or as a systemd service so it survives your SSH session ending:

tmux new -s my-crew crewai run

For production use, wrap the launch in a systemd unit that starts tmux new -d -s my-crew crewai run on boot.

What it costs

The 2 vCPU / 4 GB / 40 GB Cube described here bills at about $0.0140/hour — roughly $0.34/day, or $10/month if left running. Usage is metered by the minute.

On top of that, either an AI provider bill (paid per message) or the local Ollama path. Ollama is free to run; the only cost is the larger Cube to fit the model weights in RAM. CrewAI AMP offers a paid managed control plane at app.crewai.com — the framework on a Cube is free.

Cleaning up

Deleting the Cube removes CrewAI and any project files along with the disk. If a crew configuration matters, snapshot the Cube first.

Next steps

• Self-host OpenHands on a Cube — the control-center alternative.
• Self-host Open Interpreter on a Cube — the code-execution alternative.
• Cubes — sizing, snapshots, and how billing works when a Cube is stopped.