Quickstart: first request and Claude Code

By the end of this page you have an OpenLLM key, a working API request, and Claude Code running through OpenLLM.

Before you begin

A Google or GitHub account, or an email address, to sign in with.
A terminal with curl.
For the CLI steps, a Mac or a Linux machine. There is no Windows build.

You don't need a provider account or an API key. Every account includes a monthly allowance on hosted models, and on a new account the built-in ultra, plus, and lite chains start with those models.

Go to https://www.openllm.sh/sign-in and select Continue with Google or Continue with GitHub. To use an email address instead, select Create account and sign up with your email and a password.

You land on the vault setup screen.

Set up your vault

The vault encrypts your provider credentials and keys in your browser. Choose how this device opens it:

Where your browser supports passkeys, select Use Touch ID (or your device's own method).
Otherwise, or if you prefer, select Use a PIN instead and set a 4- or 6-digit PIN.

When the vault is ready, OpenLLM creates an API key named default and opens Get started.

You save your 12-word recovery phrase from the dashboard afterwards: Get started shows Save phrase until you do. OpenLLM can't recover the phrase for you. See Recovery phrase.

Copy your key

On Get started, the Your key row shows your default key, masked. Select the copy button next to it.

You can copy the key again at any time from Devices & keys while your vault is unlocked. Anyone who holds the key can use your connected providers, so treat it like a password. See If a key leaks.

Make your first request

Replace sk-llm-YOUR_KEY with your key and run:

curl https://www.openllm.sh/v1/chat/completions \
  -H "Authorization: Bearer sk-llm-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ultra",
    "messages": [{"role": "user", "content": "Say hello from OpenLLM."}]
  }'

You get an OpenAI-style chat completion back. ultra is a chain: OpenLLM tries its models in order until one answers. To see which model answered, add -i to the command and read the x-openllm-resolved-model response header.

Install the CLI

Run the installer with your key:

curl -fsSL https://www.openllm.sh/install | OPENLLM_API_KEY='sk-llm-YOUR_KEY' bash

The script installs openllm, starts the local daemon paired with your key, and installs any missing subscription CLIs, such as Claude Code, in the background. Open a new terminal so openllm is on your PATH, then run:

openllm status

Your machine now appears as a device in Devices & keys. For details and other options, see Install.

Run Claude Code through OpenLLM

openllm claude

Claude Code starts with OpenLLM applied to this session only. Your own Claude Code settings don't change, and the session uses ultra by default. If Claude Code isn't installed yet, wait for the background install to finish (it logs to ~/.openllm/cli-install.log) and run the command again. See Claude Code.

Check that it works

Send a one-shot prompt from the CLI. It uses ultra unless you pass -m <model>:

openllm -p "Say hello from OpenLLM."

The reply prints in your terminal.

Next steps

Connect a subscription

Use your Claude, ChatGPT, Kimi, or Grok plan through the daemon.

Set up your clients

Codex, OpenCode, Raycast, and any OpenAI or Anthropic client.

Routing and fallback

Edit ultra, plus, and lite, or build your own chains.

How it works

Where each request runs and what OpenLLM sees.

Last updated on

On this page