How it works
From an idea to a tool
you can run again.
Cargo-AI puts inputs, structured results, and action rules in a readable definition. Run it directly or turn it into a native command-line tool.
The workflow is yours
Your task. Your workflow.
Choose what you want to accomplish. Define the inputs, where AI judgment belongs, and what actions can follow. Keep precise operations in code you control.
Define the inputs.
Specify the information and context your task needs, along with the instructions the model should follow.
Choose where AI makes a judgment.
Give a supported model a defined role in your workflow. You choose what it should decide and the form of the result you need.
Check the structure in code.
Cargo-AI checks that the answer has the structure your workflow expects before running matching actions. Schema checks establish structure, not truth.
Run the actions you declare.
Define which tools can run, the conditions that trigger them, and the information they receive. Your tools perform the work and produce the outputs your task calls for.
Use the workflow again.
Run the same definition with new inputs. The instructions, checks, and action rules stay in place; the model's judgments can change.
This is targeted autonomy: the model makes a bounded judgment inside a workflow whose inputs, checks, and follow-up actions you define.
Build a workflow of your own.
Define the agent.
Describe the inputs, the structured result you need, and the actions that can follow. Your coding assistant can help turn that request into a JSON definition you can read and review.
Run and inspect.
Choose a supported model profile and run the definition locally. Cargo-AI checks the returned structure against your schema before running matching actions. Review the output and refine the workflow.
Add your own tools.
Connect project-owned Rust tools to files, APIs, or business systems. Keep calculations, validation, and other precise operations in code you can test.
Package and share.
Keep agents, tools, and assets together in a project. Run the definition directly, or hatch it into a native executable with generated Rust source you can inspect and retain.
Keep the important choices explicit.
Your definition makes the model's role and the available follow-up actions reviewable. Add task-specific checks in your own tools, and decide how you will review the results.
Choose supported hosted or local models through connection profiles. Different tasks can use different models, while the workflow and tools remain yours to maintain. When you use a hosted model, its inputs are sent to that provider.
Built on Rust. Open to inspection.
Cargo-AI is MIT-licensed open source, built on Rust and Cargo. Inspect the foundation, keep generated source, and create tools that fit your needs. Connected models and services have their own terms and capabilities.
Explore the full technical documentation on GitHub for definitions, providers, tools, and release notes.