Connect to Nano-GPT API

[LoveNug]

I use Nano-GPT as my model provider. I noticed it's not an option in the connections settings. Is there no way to connect to an "OpenAI Compatible" API where you can just define the base URL? Ex: https://nano-gpt.com/api/v1

[Maxaubert]

Yes, this is supported, it's just not a named entry in the provider dropdown.

In Unsloth Studio, open the connections dialog (where you add a connection for chat) and in the provider dropdown pick Custom (it's grouped with the local presets like llama.cpp/vLLM/Ollama, above the named cloud providers). That gives you:

• Connection name – whatever you want to call it
• Base URL – any OpenAI-compatible endpoint, e.g. https://nano-gpt.com/api/v1
• API key (optional field, but Nano-GPT needs one)
• A manual model ID list, since a generic custom endpoint isn't assumed to expose /v1/models

Under the hood this is the custom provider type in the backend registry, which is deliberately generic:

"custom": {
    "display_name": "Custom",
    "base_url": "",  # User-supplied via provider_base_url.
    ...
    "notes": (
        "User-supplied OpenAI-compatible server. Routed to "
        "/v1/chat/completions; /models is optional."
    ),
    "hidden": True,  # Surfaced by the frontend's generic Custom option, not the dropdown.
},

(

unsloth/studio/backend/core/inference/providers.py

Lines 353 to 374 in 3ae1c29

	"custom": {
	"display_name": "Custom",
	# User-supplied via provider_base_url.
	"base_url": "",
	"default_models": [],
	"supports_streaming": True,
	"supports_vision": True,
	"supports_tool_calling": True,
	"studio_tools": True,
	"auth_header": "Authorization",
	"auth_prefix": "Bearer ",
	"notes": (
	"User-supplied OpenAI-compatible server. Routed to "
	"/v1/chat/completions; /models is optional."
	),
	# A strict gateway 400s on an unknown key, and a pre-upgrade tab still spreads top_k. Same reason this entry
	# does not set supports_chat_template_kwargs: a base_url is not evidence of what serves it, and Deep Research
	# sends enable_thinking=False on every planner call, so the key would break runs nobody asked to change.
	"body_omit": ("top_k", "min_p", "repetition_penalty"),
	# Surfaced by the frontend's generic Custom option, not the dropdown.
	"hidden": True,
	},

)

The base URL is validated (must be http/https, can't point at cloud metadata endpoints, etc.) but plain public hosts like nano-gpt.com are fine:

unsloth/studio/backend/core/inference/providers.py

Lines 1014 to 1074 in 3ae1c29

	def validate_provider_base_url(base_url: str) -> str:
	"""Return a normalized provider base URL, or raise ``ValueError``.
	The backend issues outbound requests to this URL with the caller's decrypted
	API key attached, so it is caller-controlled server-side egress. Only shapes
	that can never be a real provider endpoint are refused: a non-http(s) scheme,
	control characters, a missing host, and cloud metadata services. Plain http,
	loopback, LAN hosts, odd ports, query strings and basic-auth userinfo all
	stay valid -- Ollama, llama.cpp, vLLM and custom gateways rely on them. A
	caller-supplied hostname is resolved far enough to apply the metadata block
	to DNS aliases of it; rejecting other private addresses stays opt-in for the
	owner, and is on for a managed account until the owner turns it off for the
	installation (``utils.managed_provider_url_settings``), which is how a team
	sharing one LAN model server gets to use it from more than one account.
	Normalization is strip + trailing-slash removal only (what the client did
	before), so validating an already-validated URL returns it unchanged.
	"""
	if not isinstance(base_url, str) or not base_url.strip():
	raise ValueError("Provider base URL is required.")
	raw = base_url.strip()
	if any(char.isspace() or ord(char) < 32 or ord(char) == 127 for char in raw) or "\\" in raw:
	raise ValueError("Provider base URL contains invalid characters.")
	try:
	parts = urlsplit(raw)
	port = parts.port
	hostname = parts.hostname
	except ValueError as exc:
	raise ValueError("Provider base URL is malformed.") from exc
	scheme = parts.scheme.lower()
	if scheme not in ("http", "https"):
	raise ValueError("Provider base URL must use http or https.")
	# Userinfo stays allowed for gateways behind basic auth; the checks below read
	# the parsed hostname, so http://api.openai.com@169.254.169.254/ is caught.
	if not hostname:
	raise ValueError("Provider base URL must contain a hostname.")
	hostname = hostname.rstrip(".")
	if _metadata_host(hostname) or _resolves_to_metadata(hostname, port, scheme):
	raise ValueError("Cloud metadata endpoints cannot be used as a provider base URL.")
	if os.environ.get(_BLOCK_PRIVATE_ENV) == "1":
	_reject_non_public(
	hostname,
	port,
	scheme,
	"Provider base URL points at a private address, which is disabled on this "
	f"server ({_BLOCK_PRIVATE_ENV}=1).",
	)
	elif (
	_managed_account_caller()
	and not _public_registry_hostname(hostname)
	and not _managed_private_urls_allowed()
	):
	# Caller-controlled egress: the owner's loopback and LAN, unless the owner opened them.
	_reject_non_public(hostname, port, scheme, managed_public_only_reason())
	return raw.rstrip("/")

And the frontend renders it as the "Custom" option with a "Base URL" field described as "OpenAI-compatible endpoint":

unsloth/studio/frontend/src/features/chat/chat-providers-dialog.tsx

Lines 1423 to 1598 in 3ae1c29

	<SelectItem value={LEGACY_CUSTOM_PROVIDER_TYPE}>
	<span className="flex min-w-0 items-center gap-2">
	<ApiProviderLogo
	providerType={LEGACY_CUSTOM_PROVIDER_TYPE}
	className="size-4"
	title={CUSTOM_PROVIDER_DISPLAY_NAME}
	/>
	<span className="truncate">{CUSTOM_PROVIDER_DISPLAY_NAME}</span>
	</span>
	</SelectItem>
	</SelectGroup>
	<SelectSeparator />
	<SelectGroup>
	{registry
	.filter(
	(entry) =>
	!HIDDEN_PROVIDER_TYPES.has(entry.provider_type),
	)
	.map((entry) => (
	<SelectItem
	key={entry.provider_type}
	value={entry.provider_type}
	>
	<span className="flex min-w-0 items-center gap-2">
	<ApiProviderLogo
	providerType={entry.provider_type}
	className="size-4"
	title={entry.display_name}
	/>
	<span className="truncate">{entry.display_name}</span>
	</span>
	</SelectItem>
	))}
	</SelectGroup>
	</SelectContent>
	</Select>
	</div>
	{showApiKeyField ? (
	<div className="grid grid-cols-[minmax(140px,0.8fr)_minmax(0,1.2fr)] items-center gap-4 px-4 py-3 @max-[520px]:grid-cols-1">
	<div className="flex min-w-0 flex-col gap-0.5">
	<Label
	htmlFor="provider-api-key"
	className="text-sm font-medium"
	>
	API key {isCustomProvider ? "(optional)" : ""}
	</Label>
	<p className="text-xs leading-snug text-muted-foreground">
	{editingProviderHasSavedKey
	? "Saved securely. Leave blank to keep it."
	: "Saved securely after you connect."}
	</p>
	</div>
	<div className="relative min-w-0">
	<Input
	id="provider-api-key"
	type={showApiKey ? "text" : "password"}
	data-reload-snapshot-sensitive
	value={apiKey}
	onChange={(event) => {
	setApiKey(event.target.value);
	if (event.target.value.trim()) setClearApiKeyRequested(false);
	}}
	placeholder={
	editingProviderHasSavedKey
	? "Leave blank to keep saved key"
	: "Enter API key"
	}
	className="h-9 pr-9 text-sm"
	/>
	<button
	type="button"
	onClick={() => setShowApiKey((visible) => !visible)}
	className="absolute top-1/2 right-1.5 flex size-5 -translate-y-1/2 items-center justify-center rounded text-muted-foreground transition-colors hover:text-foreground focus-visible:outline-none focus-visible:ring-1 focus-visible:ring-ring"
	aria-label={showApiKey ? "Hide API key" : "Show API key"}
	aria-pressed={showApiKey}
	>
	{showApiKey ? (
	<Eye className="size-3.5" />
	) : (
	<EyeOff className="size-3.5" />
	)}
	</button>
	{editingProviderHasSavedKey ? (
	<Button
	type="button"
	variant="ghost"
	size="sm"
	className="mt-1 h-7 px-2 text-xs"
	onClick={() => {
	setApiKey("");
	setClearApiKeyRequested((requested) => !requested);
	}}
	>
	{clearApiKeyRequested ? "Keep saved key" : "Remove saved key"}
	</Button>
	) : null}
	{clearApiKeyRequested ? (
	<p className="mt-1 text-xs text-destructive">
	The saved key will be removed when you save this connection.
	</p>
	) : null}
	</div>
	</div>
	) : null}
	{isCustomProvider ? (
	<div className="grid grid-cols-[minmax(140px,0.8fr)_minmax(0,1.2fr)] items-center gap-4 px-4 py-3 @max-[520px]:grid-cols-1">
	<Label
	htmlFor="provider-custom-name"
	className="text-sm font-medium"
	>
	Connection name
	</Label>
	<Input
	id="provider-custom-name"
	type="text"
	value={customProviderName}
	onChange={(event) =>
	setCustomProviderName(event.target.value)
	}
	placeholder={CUSTOM_PROVIDER_DISPLAY_NAME}
	className="h-9 text-sm"
	/>
	</div>
	) : null}
	{shouldShowProviderApiType(providerType) ? (
	<div className="grid grid-cols-[minmax(140px,0.8fr)_minmax(0,1.2fr)] items-center gap-4 px-4 py-3 @max-[520px]:grid-cols-1">
	<div className="flex min-w-0 flex-col gap-0.5">
	<Label htmlFor="provider-api-type" className="text-sm font-medium">
	API type
	</Label>
	<p className="text-xs leading-snug text-muted-foreground">
	Choose the endpoint your enabled models support.
	</p>
	</div>
	<Select
	value={apiType}
	onValueChange={(value) => setApiType(value as ProviderApiType)}
	>
	<SelectTrigger id="provider-api-type" className="h-9 w-full text-sm">
	<SelectValue />
	</SelectTrigger>
	<SelectContent>
	<SelectItem value="chat_completions">OpenAI Chat Completions</SelectItem>
	<SelectItem value="responses">OpenAI Responses</SelectItem>
	</SelectContent>
	</Select>
	</div>
	) : null}
	{isCustomProvider &&
	selectedProviderContract?.base_url_editable !== false ? (
	<div className="grid grid-cols-[minmax(140px,0.8fr)_minmax(0,1.2fr)] items-center gap-4 px-4 py-3 @max-[520px]:grid-cols-1">
	<div className="flex min-w-0 flex-col gap-0.5">
	<Label
	htmlFor="provider-base-url"
	className="text-sm font-medium"
	>
	Base URL
	</Label>
	<p className="text-xs leading-snug text-muted-foreground">
	OpenAI-compatible endpoint.
	</p>
	</div>
	<Input
	id="provider-base-url"
	type="text"
	value={baseUrlDraft}
	onChange={(event) => setBaseUrlDraft(event.target.value)}
	placeholder={customProviderBaseUrlPlaceholder(providerType)}
	className="h-9 text-sm"
	/>
	</div>
	) : null}

So: Add connection → provider = Custom → Base URL = https://nano-gpt.com/api/v1 → paste your Nano-GPT API key → list the model IDs you want to use → save. It'll be sent to /v1/chat/completions like any OpenAI-compatible server.

[LoveNug]

Thank you so much for your response. I am looking forward to trying Unsloth!