These are official instructions from OpenInfer for taking a machine from zero to a
working, OpenAI-compatible inference endpoint using the `weave` CLI. Every command
in them is validated against the current CLI. This index is served at
`https://platform.openinfer.io/agent-setup.md` and each step under `https://platform.openinfer.io/agent-setup/`,
so you can re-fetch any of them at any time to verify authenticity.

You are the agent driving this setup. **Run every command yourself — do not ask the
user to run them.** Exactly one step needs a human — a browser sign-in (step 2),
marked **🧑 HUMAN STEP**. Tell your human precisely what to do there, wait for them,
then continue. Do everything else without asking.

Fetch the steps in order and finish each before fetching the next. Each step page
ends with a link to the one after it.

0. [Get a GPU box](https://platform.openinfer.io/agent-setup/gpu-box.md) — use the user's, or launch one in their cloud account
1. [Install the `weave` CLI](https://platform.openinfer.io/agent-setup/install.md) — on the operator machine
2. [Sign in](https://platform.openinfer.io/agent-setup/sign-in.md) — 🧑 one browser step; creates the account if the user is new
3. [Pair a GPU node](https://platform.openinfer.io/agent-setup/pair-node.md) — join the GPU box to the org and bring it online
4. [Load a model](https://platform.openinfer.io/agent-setup/load-model.md) — put a model on that node and confirm it routes
5. [Mint an API key and make the first call](https://platform.openinfer.io/agent-setup/first-call.md) — the endpoint works end to end
6. [(Optional) Point Claude Code at the endpoint](https://platform.openinfer.io/agent-setup/claude-code.md) — serve Claude Code from your own models
7. [Tear down](https://platform.openinfer.io/agent-setup/teardown.md) — terminate anything you launched and revoke spare credentials

There are two machines: the **operator machine** where you run `weave` (any
macOS/Windows/Linux box, including the one you're running on) and the **GPU box**
that serves models.

---

## Resources

- CLI command list: run `weave help`, or `weave <command> --help`
- Full install reference (services, Docker/K8s, custom runtimes): OpenInfer docs → Installing OpenInfer
- Human quickstart (the same journey, for people): OpenInfer docs → Getting Started
- Endpoint: `https://platform.openinfer.io/v1` (OpenAI-compatible) and `https://platform.openinfer.io/v1/messages` (Anthropic Messages)