Use your API key

Your rental comes with an API key that starts with gfk_. It works exactly like an OpenAI API key, only it points at the GPU you rented. Anything that can talk to the OpenAI API can use it: the official OpenAI libraries, LangChain, most chat apps, or a plain curl command.

You need three things. You’ll find all of them on Dashboard → Current Rentals.

What	Value
Base URL (also called endpoint or API base)	https://gpuflow.app/v1
API key	your key, starting with gfk_
Model	the AI model on that GPU, for example qwen2.5:7b (step 1 shows how to find it)

In the examples below, replace gfk_your_key with your own key.

1. Check your key and find the model name

Paste this into a terminal:

curl https://gpuflow.app/v1/models \
  -H "Authorization: Bearer gfk_your_key"

You get an answer like this:

{
  "object": "list",
  "data": [{ "id": "qwen2.5:7b", "object": "model", "owned_by": "gpuflow" }]
}

The id (here qwen2.5:7b) is the model name to use. If the provider’s description lists more models, you can use those names too.

2. Send your first message

curl

curl https://gpuflow.app/v1/chat/completions \
  -H "Authorization: Bearer gfk_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen2.5:7b",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant. Answer briefly."},
      {"role": "user", "content": "Write a two-line product description for a steel water bottle."}
    ]
  }'

Python

Install the library once: pip install openai

from openai import OpenAI

client = OpenAI(
    base_url="https://gpuflow.app/v1",
    api_key="gfk_your_key",
)

reply = client.chat.completions.create(
    model="qwen2.5:7b",
    messages=[
        {"role": "system", "content": "You are a helpful assistant. Answer briefly."},
        {"role": "user", "content": "Write a two-line product description for a steel water bottle."},
    ],
)
print(reply.choices[0].message.content)

JavaScript

Install the library once: npm install openai

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://gpuflow.app/v1",
  apiKey: "gfk_your_key",
});

const reply = await client.chat.completions.create({
  model: "qwen2.5:7b",
  messages: [
    { role: "system", content: "You are a helpful assistant. Answer briefly." },
    { role: "user", content: "Write a two-line product description for a steel water bottle." },
  ],
});
console.log(reply.choices[0].message.content);

3. Get the answer word by word (streaming)

Add "stream": true to see the answer appear as it is written, like in a chat app.

curl

curl https://gpuflow.app/v1/chat/completions \
  -H "Authorization: Bearer gfk_your_key" \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen2.5:7b", "stream": true,
       "messages": [{"role": "user", "content": "Explain what a GPU is in three sentences."}]}'

Python

stream = client.chat.completions.create(
    model="qwen2.5:7b",
    stream=True,
    messages=[{"role": "user", "content": "Explain what a GPU is in three sentences."}],
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="", flush=True)

Use it in an app

Many apps have a setting called “OpenAI-compatible”, “Custom OpenAI” or “OpenAI API base”. Fill it in like this:

1. Base URL / API base / endpoint: https://gpuflow.app/v1
2. API key: your gfk_… key
3. Model: the name from step 1, for example qwen2.5:7b

If the app asks for a “provider”, pick “OpenAI” or “OpenAI-compatible”.

What the key can and can’t do

• It works for chat: /v1/models and /v1/chat/completions.
• It works until your rental ends. After that it answers with invalid_api_key.
• It is yours alone. Anyone who has it can use your rental, so don’t paste it into public code or chats. If it leaks, click New key on Current Rentals. The old key stops working at once.

Your prompts run on someone's own computer

The AI model runs on the provider’s machine, so your prompts and the answers pass through it. Don’t send passwords, card numbers or other secrets you wouldn’t share with a stranger.

Something not working? See fix a problem.