Open-source AI agent builder

Build AI tools.
Keep them yours.

Create agents you can run, inspect, and share—with the models you choose.

Build with Codex or Claude Code.

Cargo-AI runtime · MIT licensed · macOS · Linux · Windows

The workflow is yours

AI judgment.
A flow you define.
  1. Inputs you defineThe information and context for your task
  2. AI judgment where you need itYour chosen model, with a role you define
  3. Code checks the structureField types and allowed values
  4. Actions you declareYour tools, outputs, and action rules
  5. A workflow you can run againNew inputs, the same defined process

You define the task and action rules. AI makes the judgment within them.

Own what you build.

Keep your agent definitions and tool source. Inspect, adapt, and share your work.

Choose what powers it.

Use supported local or hosted models. Match the model to the task, and change it as your needs evolve.

Define how it acts.

Make inputs, outputs, and action rules explicit in a readable agent definition.

New · Desktop for macOS

Cargo-AI Desktop

A desktop home for your Cargo-AI work.

Preview version · May contain bugs

Explore Desktop

Start with one useful task

What would you like to build?

Build your first agent