OpenAI SDK Quickstart (DOCS-01)

Point any OpenAI-compatible client at your Yaarbal capsule and get free inference from a real GPU on the network. No new SDK to learn — the capsule speaks the real POST /v1/chat/completions (streaming included) and GET /v1/models contract (rentmygpu/seeder/capsule_routes.py), so this works with openai-python, aider, Continue, or anything else that already speaks OpenAI's API shape.

Your invite includes your base_url and api_key

There is no self-serve signup or token-minting page in this beta. When you're invited, the operator sends you your capsule's base_url (already pointed at its /v1 surface) and your rmgc- API key together, bundled with the invite. Drop both straight into the snippet below — there's nothing else to provision.

Copy-paste

from openai import OpenAI

client = OpenAI(base_url="<base_url-from-your-invite>", api_key="rmgc-<from-your-invite>")
resp = client.chat.completions.create(
    model="<model-name-from-your-invite>",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

That's the whole integration. model is whatever your invite tells you the capsule is serving — capsules run one curated model at a time.

Notes

Operator appendix (not the user path)

This section is for the operator provisioning access, not for someone following the quickstart above. Mint a token with:

python -m rentmygpu.cli.capsule_tokens add --peer-id <peer-id> --label <name>

This prints the raw rmgc- token to your terminal exactly once — send it, together with the capsule's base_url, to the teammate you're inviting. See docs/CAPSULE-SECURITY.md for the full trust-boundary and token-handling contract before handing out a token.