An OpenAI-compatible API on a node you run

Two ways to get a model

Make an API key

1. Open the model's page. On your arcade, open Exchange → LLMs and tap the model's card.
2. Make a key. Under Use it from your own apps, press Make a key, give it a name and pick which seller and coin it uses.
3. Copy it now. The key starts with sk-arcade- and is shown once. The arcade keeps only a scrambled copy, and you can revoke the key at any time.
4. Use your arcade as the base URL. https://your-arcade.example/v1, with the key as the bearer token.

curl https://your-arcade.example/v1/chat/completions \
  -H "Authorization: Bearer $ARCADE_API_KEY" -H "Content-Type: application/json" \
  -d '{"model": "gpt-oss-20b", "messages": [{"role": "user", "content": "Hello"}]}'

Ask /v1/models for the exact model ids your key can use.

The OpenAI SDK, with only base_url changed

Python

from openai import OpenAI

client = OpenAI(base_url="https://your-arcade.example/v1",
                api_key="sk-arcade-...")

print([m.id for m in client.models.list()])

reply = client.chat.completions.create(
    model="gpt-oss-20b",
    messages=[{"role": "user", "content": "Name three Game Boy colours."}],
)
print(reply.choices[0].message.content)

# Streaming works the same way
for chunk in client.chat.completions.create(model="gpt-oss-20b", stream=True,
        messages=[{"role": "user", "content": "Write a haiku about blocks."}]):
    print(chunk.choices[0].delta.content or "", end="")

JavaScript (Node.js)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://your-arcade.example/v1",
  apiKey: process.env.ARCADE_API_KEY,
});

const reply = await client.chat.completions.create({
  model: "gpt-oss-20b",
  messages: [{ role: "user", content: "Name three Game Boy colours." }],
});
console.log(reply.choices[0].message.content);

No code change at all

Both SDKs read the base URL and key from the environment, so many tools that use them switch over with two lines:

export OPENAI_BASE_URL=https://your-arcade.example/v1
export OPENAI_API_KEY=sk-arcade-...

Calling a model from an inscribed game

A game on the Apps tab holds no key and runs no server. It loads /r/llm.js, and the player is asked once whether the game may use a model. The request takes the same fields as a chat completion.

<script src="/r/llm.js"></script>
<script>
arcade.llm.chat({messages: [{role: 'user', content: 'Name a dragon'}],
                 onToken: function (text) { out.textContent += text; }})
  .then(function (r) { /* r.content is the whole reply */ },
        function (e) { /* e.message: "declined", "sign in first" ... */ });
</script>

arcade.llm.image({prompt, size}) makes a picture the same way, and arcade.llm.offers() lists the models on offer. See publishing a game and the inscription API.