Quickstart

The goal here is one working request, then your first tool wired up. The gateway lives at openllm.sh; swap in sk-llm-YOUR_KEY with your own key as you go.

On first sign-in, an onboarding wizard walks you through most of this — the vault, your first key, your device, and setup options — so in practice you follow the screens and land at step 5 with everything in place. The steps below explain what each screen is doing, and double as the manual path if you skipped onboarding or are adding things later.

1. Sign in

Go to openllm.sh and continue with Google. That is your account. There is nothing to install yet.

2. Get your encrypted vault and save the recovery phrase

Onboarding's first stop is your vault, the encrypted store that holds every provider credential you will connect. It sets itself up automatically; the one thing you have to do is save the recovery phrase it shows you.

That phrase is the master secret behind the vault. It is what your credentials are ultimately encrypted under, and OpenLLM does not keep a copy of it. That is the point (a database dump reveals nothing usable without it), but it also means:

• Store it somewhere safe now (a password manager is ideal). The backup is taken once, here, and the step asks you to type a few of the words back before it lets you continue.
• If you lose it, it cannot be recovered. Your only move is to reset the vault, which re-keys it from scratch: every stored provider credential is destroyed, every existing sk-llm-… key is invalidated, and saved memories encrypted under the old phrase go with them.

Think of the recovery phrase the way you would a wallet seed phrase: whoever has it can unlock the vault, and no one (including OpenLLM) can restore it for you. See Security for the full model.

3. Get your OpenLLM key

Onboarding mints your first key for you. It looks like sk-llm-…. Copy it — this single key is what every tool will use. (Later, you can create more keys under Devices & keys; one per machine is the pattern.)

4. Connect your device

Next, onboarding pairs your machine. It hands you an install command with your key already in it; run it on your machine (the same flow lives under Devices & keys → Add a device afterwards):

curl -fsSL "https://openllm.sh/api/setup/daemon/install.sh" \
  | OPENLLM_API_KEY="sk-llm-YOUR_KEY" bash

That installs the local daemon and pairs the machine to your account — the wizard flips to Connected on its own. The daemon is what runs subscription providers on your own machine, and it keeps itself running across reboots. See Use multiple devices for the full model.

5. Connect a provider

With onboarding done, you land on the dashboard. Go to Providers and connect at least one. You have two ways to do it:

• Paste an API key for Anthropic, OpenAI, Google Gemini, Alibaba DashScope, or AWS Bedrock (plus custom OpenAI-compatible endpoints).
• Connect an account over OAuth for your Claude, ChatGPT, Kimi, or Grok subscription, which runs through the daemon you just installed.

Whatever you connect goes into the vault as ciphertext. See Security for how that works.

6. Make your first request

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

lite, plus, and ultra are the preset tier aliases; you can rename them or add your own. OpenLLM resolves each to a real provider and model through your fallback chain, so you can reorder providers without touching your code. See Concepts and Fallback chains.

Anthropic-style clients can call /v1/messages instead. Embeddings, images, and audio live behind /v1/embeddings, /v1/images/generations, and /v1/audio/*.

7. Wire up a coding tool

Go to Integrations in the dashboard, pick your client, and hit install. Because your device is already paired, the install runs through the daemon with one click — no terminal needed.

Prefer to run it yourself? Each tool also has a one-line installer that bakes in your gateway URL and key. Here is Claude Code:

curl -fsSL "https://openllm.sh/api/setup/claude-code/install.sh" \
  | OPENLLM_API_KEY="sk-llm-YOUR_KEY" bash

Either way, the install merges into ~/.claude/settings.json without clobbering your existing hooks or permissions, points ANTHROPIC_BASE_URL at your gateway, and maps Opus, Sonnet, and Haiku to the ultra, plus, and lite tiers. Restart Claude Code and it routes through OpenLLM.

Full instructions for every tool, including Codex, the Kimi CLI, and Raycast, are on Connect your tools.

You are done

Every tool now runs on your accounts through one key. Good next steps:

• Set up Fallback chains so a task never stalls when one provider is down.
• Read Why it is compliant to understand what OpenLLM does and does not do with your accounts.