Getting started

Install the Auremi CLI and skill in your repo, then let your agent set up the model and start training.

You run one install command in your repo. After that, you ask your coding agent (Claude Code, Codex, or Cursor) for what you want, and it runs the Auremi CLI for you. Under each prompt you can open What the agent runs to see the commands.

1. Install the CLI and skill

   From your repo’s folder, run the command for your agent. It installs the auremi CLI from npm and adds the Auremi skill to the repo. You need Node.js 22 or newer.

   Claude Code

   curl -fsSL https://auremi.ai/install.sh | AUREMI_AGENT=claude sh
   

   To install the CLI on its own, run npm install -g @auremi/cli.

2. Ask your agent to set up Auremi

   Set up the repo

   CopiedCopy prompt

   The agent creates a starter model with an example notebook, checks it, and pushes the notebook. To sign in, it asks you to run auremi login yourself and approve the link in your browser. It never asks for your credentials in chat.

   What the agent runs

   auremi login   # you run this one
   auremi init --name my-model
   auremi check
   auremi notebook push --message "First notebook"
   
3. Start a training run

   Train the model

   CopiedCopy prompt

   Training needs a dataset. The CLI won’t pick one for you, so the agent asks which one to use. Training runs from your latest commit on GitHub, so the agent commits and pushes first.

   What the agent runs

   git add -A
   git commit -m "Starter model"
   git push
   auremi dataset list
   auremi train --dataset <datasetVersionId> --no-run
   auremi train --dataset <datasetVersionId> --wait
   

Training runs on a training worker or on your own cloud account. To set one up, see Training workers.

For every command, see CLI commands.