Set up your coding tools

npx uberllm init

The command finds the coding tools on your machine and sets each one up to use UberLLM. You need Node.js 18 or later. Nothing else is installed.

What happens when you run it:

1. It lists the tools it found.
2. Your browser opens. Sign in (or create an account) and press Connect. This makes an API key for this machine.
3. The key is saved to ~/.uberllm/key, a file only you can read.
4. Each tool's own config file is changed to use UberLLM. A copy of every file is saved first.
5. It checks that the key works and tells you what to restart.

Running it again changes nothing if everything is already set up.

See the changes first

npx uberllm init --dry-run

This prints each file it would change and the lines it would add. It does not open the browser and does not write anything.

What each tool gets

Kilo Code and Goose only read a key from the environment. For them, one line is added to your shell profile (~/.zshrc, ~/.bashrc or fish). That line reads the key file; the key itself is not written into the profile. On Windows the command prints that one step for you to do instead.

Claude Code keeps asking for its own Claude models. The other tools get a default model: the first coding model on the model list that is taking requests when you run the command. To choose it yourself:

npx uberllm init --model deepseek/deepseek-v4.1-flash

Tools that are set up inside the app

Cursor, Cline and Roo Code keep these settings inside the editor, not in a file a command can change. For them the command prints the three steps to follow, with the address and model id filled in. See Cursor, Cline and Roo Code.

Gemini CLI only talks to servers that speak Google's Gemini API, which UberLLM does not serve. To use Gemini models through UberLLM, set up OpenCode or Codex with a Gemini model id: npx uberllm init --tool opencode --model <id>.

Options

Which key is used: --key if you pass it, then the key already saved in ~/.uberllm, then UBERLLM_API_KEY from the environment, then the browser.

One project only

cd my-project
npx uberllm init --project

Only that folder runs on UberLLM; your other projects are unchanged. This works for the tools that have a per-project config: Claude Code (.claude/settings.local.json), OpenCode (opencode.json) and Aider (.aider.conf.yml). No key is written into the project folder. Codex CLI, Kilo Code, Goose and Continue are set up per machine, so they are skipped here.

On Windows, Claude Code is not set up per project, because that would put the key in the project folder. Run the command without --project there.

Undo

npx uberllm init --undo

• A file you have not touched since is put back exactly as it was. A file the command created is removed.
• A file you have edited since keeps your edits. Only what the command added is taken out, and the settings it replaced are put back. The copy from before the command ran stays in ~/.uberllm/backups/.
• The saved key file is removed if this command created it and no tool it set up still uses it. A key file from the Claude Code installer is left in place. To switch the key off as well, delete it on your keys page.

Undo one tool with --undo --tool codex, or one project with --undo --project. Kilo Code and Goose share the line in your shell profile, so it stays until both are undone.

Where things are kept

The key is sent to api.uberllm.dev and nowhere else. It is never printed.

When a file is left alone

The command does not rewrite a file it could damage. It tells you and prints the steps to do by hand. This happens when:

• a JSON config has comments (opencode.jsonc, kilo.jsonc), because a rewrite would lose them;
• a JSON config cannot be read;
• Continue's models: is written on one line;
• with --project, a config file is a link to a file outside the project.

The tool is then listed as "not set up yet" at the end of the run.

Troubleshooting