Your AI agent build tool.

cargo-ai TM

Own your agents. Create custom tools. Share anywhere.

  • Create an agent with Codex or your favorite editor.
  • Give your agents their own custom tools.
  • Run locally and iterate quickly.
  • Share across macOS, Linux, and Windows.
View GitHub Install via Cargo Watch Demo

Open source, MIT licensed, local-first, and built for auditability.

Single JSON definition

{
  "inputs": [...],
  "agent_schema": {...},
  "actions": [...]
}

Run locally

cargo ai run agent_x.json

Share anywhere

macOS, Linux, Windows

Open source, MIT licensed, local-first, and built for auditability.

Why Cargo-AI

Real Inputs, Structured Results

Works with text, URL, image, and file inputs.

Use Your Existing ChatGPT Plan

Run with your Codex login, or bring your own API key or local model.

The Power to Create Your Own Tools

Create advanced Rust tools and integrate them directly with your agents.

Built on Your Machine

Run locally with your own profiles and keep control of the files, tools, and outputs.

Inspectable by Design

Keep agent behavior in JSON your team can read, diff, review, and improve.

Package Real Projects

Keep agents, custom tools, and assets together when a workflow grows beyond one file.

Choose the model path that fits your workflow.

GPT-5, or local Ollama models like Gemma, Llama, and Mistral.

How It Fits Together

1. Define the Agent

Start with a small JSON definition for inputs, structured output, and allowed actions.

2. Run and Inspect

Run the JSON locally, inspect the result, and iterate before exporting anything.

3. Add or Create Your Own Tools

Create advanced Rust tools and make them part of the agent workflow.

4. Package and Share

Assemble agents, tools, and assets into a project that can move across machines.

Quick Start

Before you start, install Rust and Cargo, Codex CLI, and the Codex app. This guide then installs Cargo-AI, connects your ChatGPT account through Codex CLI, and opens your first project in the Codex app.

First local setup

For the best authoring experience, use the Codex app. Codex CLI handles sign-in and remains available when you want a terminal workflow.

Setup commands

  1. Install Cargo-AIcargo install cargo-ai --locked
  2. Use Codex CLI as the auth bridgecodex login
  3. Add the default OpenAI profilecargo ai profile add openai-account --server openai --model gpt-5.4 --auth openai_account --default
  4. Authenticate that profilecargo ai auth login openai --profile openai-account
  5. Verify the profilecargo ai run adder_test --profile openai-account
  6. Create your first projectcargo ai new my-agent
  7. Enter the projectcd my-agent
  8. Add Codex guidancecargo ai add guidance --style codex
  9. Open the project in Codexopen -a Codex .

Start building

Now ask Codex what to build

Describe the workflow, inputs, outputs, and any files or APIs the agent should use.

Hey Codex, build me a Cargo-AI agent with any supporting Cargo-AI tools that can ...

Codex/OpenAI sign-in is separate from any Cargo-AI account. This first run stays local; account-backed publishing and sharing come later.

Ready for a Deeper Dive?

Use GitHub for the full developer path: README, demos, examples, install notes, tool authoring guidance, and release history.

Continue on GitHub

Explore cargo-ai on GitHub