Skip to main content

First steps

This section assumes that you have:

  • an engine and project are created and ready on the AIchor UI
  • that your account has been added to the project as a user

After installing the AIchor CLI, the next steps are:

Starting local repo from scratch​

  • If local code doesnt exist and the project will be started from scratch, you can use:
aichor local-repo init

This will automatically generate a basic Dockerfile, Pyproject.toml and AIchor manifest.yaml, as well as a directory tree compliant with uv.
The project will then be ready to run experiments.

Setting up CLI context​

  • If a local repo already exists, then the default context should be set on AIchor to simplify running requests
aichor context set project <project-name>
aichor context set engine <engine-name>
Not sure which project or engine to use?

The List Projects and Engines example script lists all available projects and engines and sets the CLI context interactively.

Login to AIchor with CLI​

An API Key can be obtained from https://web.aichor.ai/credentials

Login with:

aichor auth key

or, if you dont mind having your key in the terminal history:

aichor auth key --apikey <key>

Submitting first experiment​

For submitting the first experiment you can use the local code with:

aichor experiments submit local --message "AIchor CLI: First experiment" --repo-dir .

or use a previous commit from your remote repo with:

aichor experiment submit commit-sha d5dcb694680dbe4f26d0755eed72ec8ea336c3c2 --branch main
Submit, monitor and get notified in one go

The Submit and Monitor example script submits an experiment, streams its logs until it finishes, and sends a desktop notification (macOS) or terminal bell on completion or failure — all in one command.

Next Steps​

For an overview of all commands please check the Commands page.

For ready-to-use bash scripts covering common workflows (submitting experiments, storage operations, multi-manifest runs), see the Example Scripts page.

To manage experiments and storage directly from your IDE using natural language, see the Claude Skills page.