Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
"""Sandbox Low (python/terminal without the OS sandbox) is the Unsloth UI's to pick, like full access."""
import pytest
from fastapi import FastAPI, HTTPException
from fastapi.testclient import TestClient
from auth.authentication import get_current_subject
from routes import inference
from state.tool_policy import require_tool_access
from utils.account_context import OWNER, as_api_caller, bind_account, reset_account
_OPT_IN = "UNSLOTH_STUDIO_ALLOW_API_FULL_ACCESS"
@pytest.fixture(autouse = True)
def no_opt_in(monkeypatch):
monkeypatch.delenv(_OPT_IN, raising = False)
def _under(account, *args, **kwargs):
token = bind_account(account)
try:
require_tool_access(*args, **kwargs)
finally:
reset_account(token)
@pytest.mark.parametrize("mode", [None, "auto", "ask", "off"])
def test_sandbox_level_low_is_refused_for_api_callers(mode):
with pytest.raises(HTTPException) as refused:
_under(as_api_caller(OWNER), mode, sandbox_level = "low")
assert refused.value.status_code == 400
assert _OPT_IN in refused.value.detail
_under(as_api_caller(OWNER), mode, sandbox_level = "high")
_under(as_api_caller(OWNER), mode)
@pytest.mark.parametrize("level", ["low", "high", None])
def test_owner_ui_session_keeps_sandbox_low(level):
_under(OWNER, "auto", sandbox_level = level)
def test_operator_opt_in_restores_api_sandbox_low(monkeypatch):
monkeypatch.setenv(_OPT_IN, "1")
_under(as_api_caller(OWNER), "auto", sandbox_level = "low")
def _client_for(account):
app = FastAPI()
async def subject():
token = bind_account(account)
try:
yield account.username
finally:
reset_account(token)
app.dependency_overrides[get_current_subject] = subject
app.include_router(inference.router, prefix = "/v1")
return TestClient(app)
_REQUESTS = {
"/v1/chat/completions": {"model": "m", "messages": [{"role": "user", "content": "hi"}]},
"/v1/messages": {
"model": "m",
"max_tokens": 8,
"messages": [{"role": "user", "content": "hi"}],
},
}
@pytest.mark.parametrize("path", sorted(_REQUESTS))
@pytest.mark.parametrize("api", [True, False], ids = ["api_key", "ui_session"])
def test_the_door_refuses_only_api_key_sandbox_low(monkeypatch, path, api):
def backend(*_args, **_kwargs):
raise HTTPException(status_code = 503, detail = "no backend in this test")
monkeypatch.setattr(inference, "get_llama_cpp_backend", backend)
monkeypatch.setattr(inference, "produce_openai_chat_completions", backend)
body = {**_REQUESTS[path], "sandbox_level": "low"}
with _client_for(as_api_caller(OWNER) if api else OWNER) as client:
response = client.post(path, json = body)
if api:
assert response.status_code == 400, response.text
assert _OPT_IN in response.json()["detail"]
else:
assert response.status_code == 503, response.text
@pytest.mark.parametrize("api", [True, False], ids = ["api_key", "ui_session"])
def test_durable_chat_runs_refuse_api_key_sandbox_low_before_queueing(api):
from routes import chat_generation_runs as runs
payload = runs.CreateChatGenerationRun(
runId = "r1",
threadId = "t1",
userMessageId = "u1",
assistantMessageId = "a1",
requestPayload = {"messages": [{"role": "user", "content": "hi"}], "sandbox_level": "low"},
)
token = bind_account(as_api_caller(OWNER) if api else OWNER)
try:
if api:
with pytest.raises(HTTPException) as refused:
runs._sanitize_request(payload)
assert refused.value.status_code == 400 and _OPT_IN in refused.value.detail
else:
runs._sanitize_request(payload)
finally:
reset_account(token)