Unsloth Studio

This launcher clones the unslothai/unsloth repository and runs Unsloth Studio locally through Pinokio.

What It Does

• Clones the upstream repository into app/
• Builds the Studio frontend from studio/frontend
• Installs the Studio Python stack into a local Pinokio venv at env/
• Installs and validates upstream prebuilt llama.cpp under app/.unsloth/llama.cpp for GGUF tooling support
• Starts the Studio backend on 127.0.0.1 with an auto-assigned port
• Keeps Unsloth Studio runtime data under app/.unsloth/

How To Use

1. Click Install.
2. Click Start.
3. On first launch, open the Terminal tab and copy the bootstrap password printed by the backend.
4. Sign in with username unsloth and the printed password.
5. Change the password when prompted.

The first install or update may still take longer than a normal Python-only setup because the launcher provisions GGUF tooling. It uses Unsloth Studio's upstream prebuilt llama.cpp installer and validates llama-server before activation.

The launcher keeps the cloned repo in app/. Reset removes the cloned repo and any launcher-created venv state.

API

Most API routes require a bearer token from the auth endpoints below. Health and auth-status are available without login.

Replace 127.0.0.1:8000 below with the actual Studio URL shown by the launcher, for example the Open Studio tab or the Terminal output after startup.

Curl

Check the server:

curl http://127.0.0.1:PORT/api/health

Check auth bootstrap state:

curl http://127.0.0.1:PORT/api/auth/status

Log in and get tokens:

curl -X POST http://127.0.0.1:PORT/api/auth/login \
  -H "Content-Type: application/json" \
  -d '{"username":"unsloth","password":"YOUR_BOOTSTRAP_PASSWORD"}'

Query the current inference status:

curl http://127.0.0.1:PORT/api/inference/status \
  -H "Authorization: Bearer YOUR_ACCESS_TOKEN"

OpenAI-compatible chat after a model is loaded:

curl -X POST http://127.0.0.1:PORT/v1/chat/completions \
  -H "Authorization: Bearer YOUR_ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "active-model",
    "messages": [
      {"role": "user", "content": "Hello from Pinokio"}
    ],
    "stream": false
  }'

Python

import requests

base_url = "http://127.0.0.1:PORT"

health = requests.get(f"{base_url}/api/health", timeout=30)
print(health.json())

login = requests.post(
    f"{base_url}/api/auth/login",
    json={
        "username": "unsloth",
        "password": "YOUR_BOOTSTRAP_PASSWORD",
    },
    timeout=30,
)
token = login.json()["access_token"]

status = requests.get(
    f"{base_url}/api/inference/status",
    headers={"Authorization": f"Bearer {token}"},
    timeout=30,
)
print(status.json())

JavaScript

const baseUrl = "http://127.0.0.1:PORT";

const health = await fetch(`${baseUrl}/api/health`);
console.log(await health.json());

const login = await fetch(`${baseUrl}/api/auth/login`, {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
    username: "unsloth",
    password: "YOUR_BOOTSTRAP_PASSWORD",
  }),
});

const { access_token } = await login.json();

const status = await fetch(`${baseUrl}/api/inference/status`, {
  headers: {
    Authorization: `Bearer ${access_token}`,
  },
});

console.log(await status.json());