feat: allow overriding base URL of OpenAI models (#368)

* Add OpenAI-compatible API option in model configuration

Backend-only cherry-pick from 20617db37a.

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* fix: harden the base url settings in SaaS mode

---------

Co-authored-by: Chris Briddock <briddockchristopher@gmail.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

api/services/configuration/check_validity.py

@@ -13,6 +13,7 @@ from api.schemas.user_configuration import (
 )
 from api.services.configuration.registry import ServiceConfig, ServiceProviders
 from api.services.mps_service_key_client import mps_service_key_client
+from api.utils.url_security import validate_user_configured_service_url
 
 AuthContext = TypedDict(
     "AuthContext",
@@ -107,6 +108,17 @@ class UserConfigurationValidator:
 
         provider = service_config.provider
 
+        for url_field in ("base_url", "endpoint"):
+            url = getattr(service_config, url_field, None)
+            if url:
+                try:
+                    validate_user_configured_service_url(
+                        url,
+                        field_name=url_field,
+                    )
+                except ValueError as e:
+                    return [{"model": service_name, "message": str(e)}]
+
         # Speaches doesn't require an API key
         if provider == ServiceProviders.SPEACHES.value:
             try:
@@ -181,30 +193,92 @@ class UserConfigurationValidator:
         api_key = service_config.api_key
 
         try:
-            if not self._check_api_key(provider, api_key):
+            if not self._check_api_key(provider, api_key, service_config):
                 return [
-                    {"model": service_name, "message": f"Invalid {provider} API key"}
+                    {
+                        "model": service_name,
+                        "message": (
+                            f"Invalid {provider} API key. Please verify your API key is "
+                            f"correct, has not expired, and has the required permissions."
+                        ),
+                    }
                 ]
         except ValueError as e:
             return [{"model": service_name, "message": str(e)}]
 
         return []
 
-    def _check_api_key(self, provider: str, api_key: str) -> bool:
+    def _check_api_key(
+        self,
+        provider: str,
+        api_key: str,
+        service_config: Optional[ServiceConfig] = None,
+    ) -> bool:
         """Check if an API key for a provider is valid."""
         validator = self._validator_map.get(provider)
         if not validator:
             return False
 
+        if provider in (
+            ServiceProviders.OPENAI.value,
+            ServiceProviders.OPENAI_REALTIME.value,
+        ):
+            return validator(provider, api_key, service_config)
         return validator(provider, api_key)
 
-    def _check_openai_api_key(self, model: str, api_key: str) -> bool:
-        client = openai.OpenAI(api_key=api_key)
+    def _check_openai_api_key(
+        self, model: str, api_key: str, service_config: Optional[ServiceConfig] = None
+    ) -> bool:
+        client_kwargs: dict[str, str] = {"api_key": api_key}
+        base_url = getattr(service_config, "base_url", None) if service_config else None
+        if base_url:
+            client_kwargs["base_url"] = base_url
+        client = openai.OpenAI(**client_kwargs)
         try:
             client.models.list()
             return True
         except openai.AuthenticationError:
-            return False
+            if base_url and "openai.com" not in base_url:
+                raise ValueError(
+                    f"Invalid OpenAI API key. The key was rejected by the API at {base_url}. "
+                    "Please check that your API key is correct and has not been revoked."
+                )
+            raise ValueError(
+                "Invalid OpenAI API key. The key was rejected by the OpenAI API. "
+                "Please check that your API key is correct and has not been revoked. "
+                "You can verify your keys at https://platform.openai.com/api-keys."
+            )
+        except openai.APIConnectionError:
+            if base_url:
+                raise ValueError(
+                    f"Could not connect to the OpenAI-compatible API at {base_url}. "
+                    "Please verify that the base_url is correct and reachable, and try again."
+                )
+            raise ValueError(
+                "Could not connect to the OpenAI API. Please check your network connection "
+                "and try again."
+            )
+        except openai.APIError:
+            if base_url:
+                raise ValueError(
+                    f"The OpenAI-compatible API at {base_url} returned an error while "
+                    "validating the API key. Please verify that the base_url is correct, "
+                    "the service is available, and the API key is valid."
+                )
+            raise ValueError(
+                "The OpenAI API returned an error while validating the API key. "
+                "Please try again later."
+            )
+        except Exception:
+            if base_url:
+                raise ValueError(
+                    f"Failed to validate the OpenAI API key using the API at {base_url}. "
+                    "Please verify that the base_url is correct and reachable, and that the "
+                    "API key is valid."
+                )
+            raise ValueError(
+                "Failed to validate the OpenAI API key. Please try again later."
+            )
 
     def _check_deepgram_api_key(self, model: str, api_key: str) -> bool:
         try:
@@ -212,7 +286,11 @@ class UserConfigurationValidator:
             deepgram.manage.v1.projects.list()
             return True
         except Exception:
-            return False
+            raise ValueError(
+                "Invalid Deepgram API key. The key was rejected by the Deepgram API. "
+                "Please check that your API key is correct and active. "
+                "You can verify your keys at https://console.deepgram.com/."
+            )
 
     def _check_groq_api_key(self, model: str, api_key: str) -> bool:
         client = Groq(api_key=api_key)
@@ -220,7 +298,11 @@ class UserConfigurationValidator:
             client.models.list()
             return True
         except Exception:
-            return False
+            raise ValueError(
+                "Invalid Groq API key. The key was rejected by the Groq API. "
+                "Please check that your API key is correct and active. "
+                "You can verify your keys at https://console.groq.com/keys."
+            )
 
     def _validate_elevenlabs_api_key(self, model: str, api_key: str) -> bool:
         return True

api/services/configuration/registry.py

@@ -290,6 +290,10 @@ class OpenAILLMService(BaseLLMConfiguration):
         description="OpenAI chat model to use.",
         json_schema_extra={"examples": OPENAI_MODELS, "allow_custom_input": True},
     )
+    base_url: str = Field(
+        default="https://api.openai.com/v1",
+        description="Override only if using an OpenAI-compatible API (e.g. local LLM, proxy).",
+    )
 
 
 @register_llm

api/services/gen_ai/embedding/openai_service.py

@@ -11,6 +11,7 @@ from loguru import logger
 from openai import AsyncOpenAI
 
 from api.db.db_client import DBClient
+from api.utils.url_security import validate_user_configured_service_url
 
 from .base import BaseEmbeddingService
 
@@ -54,6 +55,10 @@ class OpenAIEmbeddingService(BaseEmbeddingService):
         if self._api_key_configured:
             client_kwargs = {"api_key": api_key}
             if base_url:
+                validate_user_configured_service_url(
+                    base_url,
+                    field_name="base_url",
+                )
                 client_kwargs["base_url"] = base_url
             self.client = AsyncOpenAI(**client_kwargs)
             logger.info(f"OpenAI embedding service initialized with model: {model_id}")

api/services/pipecat/service_factory.py

@@ -7,6 +7,7 @@ from loguru import logger
 from api.constants import MPS_API_URL
 from api.services.configuration.registry import ServiceProviders
 from api.services.pipecat.minimax_tts import MiniMaxOwnedSessionTTSService
+from api.utils.url_security import validate_user_configured_service_url
 from pipecat.services.assemblyai.stt import AssemblyAISTTService, AssemblyAISTTSettings
 from pipecat.services.aws.llm import AWSBedrockLLMService, AWSBedrockLLMSettings
 from pipecat.services.azure.llm import AzureLLMService, AzureLLMSettings
@@ -62,6 +63,16 @@ if TYPE_CHECKING:
     from api.services.pipecat.audio_config import AudioConfig
 
 
+def _validate_runtime_service_url(url: str, field_name: str) -> None:
+    try:
+        validate_user_configured_service_url(
+            url,
+            field_name=field_name,
+        )
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e)) from e
+
+
 def create_stt_service(
     user_config, audio_config: "AudioConfig", keyterms: list[str] | None = None
 ):
@@ -174,6 +185,7 @@ def create_stt_service(
         )
     elif user_config.stt.provider == ServiceProviders.SPEACHES.value:
         language = getattr(user_config.stt, "language", None)
+        _validate_runtime_service_url(user_config.stt.base_url, "base_url")
         return SpeachesSTTService(
             base_url=user_config.stt.base_url,
             api_key=user_config.stt.api_key or "none",
@@ -301,6 +313,7 @@ def create_tts_service(user_config, audio_config: "AudioConfig"):
         # ElevenLabs TTS uses WebSocket. Users configure base_url with an HTTP
         # scheme (matching ElevenLabs documentation, e.g.
         # https://api.eu.residency.elevenlabs.io); rewrite it to the WS scheme.
+        _validate_runtime_service_url(user_config.tts.base_url, "base_url")
         elevenlabs_url = user_config.tts.base_url.replace("https://", "wss://").replace(
             "http://", "ws://"
         )
@@ -376,6 +389,7 @@ def create_tts_service(user_config, audio_config: "AudioConfig"):
         tts._settings.language = language
         return tts
     elif user_config.tts.provider == ServiceProviders.SPEACHES.value:
+        _validate_runtime_service_url(user_config.tts.base_url, "base_url")
         return SpeachesTTSService(
             base_url=user_config.tts.base_url,
             api_key=user_config.tts.api_key or "none",
@@ -461,6 +475,7 @@ def create_tts_service(user_config, audio_config: "AudioConfig"):
         ).rstrip("/")
         if not base_url.endswith("/t2a_v2"):
             base_url = f"{base_url}/t2a_v2"
+        _validate_runtime_service_url(base_url, "base_url")
 
         session = aiohttp.ClientSession()
         return MiniMaxOwnedSessionTTSService(
@@ -504,6 +519,10 @@ def create_llm_service_from_provider(
     """
     logger.info(f"Creating LLM service: provider={provider}, model={model}")
     if provider == ServiceProviders.OPENAI.value:
+        kwargs = {}
+        if base_url:
+            _validate_runtime_service_url(base_url, "base_url")
+            kwargs["base_url"] = base_url
         if "gpt-5" in model:
             return OpenAILLMService(
                 api_key=api_key,
@@ -511,10 +530,12 @@ def create_llm_service_from_provider(
                     model=model,
                     extra={"reasoning_effort": "minimal", "verbosity": "low"},
                 ),
+                **kwargs,
             )
         return OpenAILLMService(
             api_key=api_key,
             settings=OpenAILLMSettings(model=model, temperature=0.1),
+            **kwargs,
         )
     elif provider == ServiceProviders.GROQ.value:
         return GroqLLMService(
@@ -524,6 +545,7 @@ def create_llm_service_from_provider(
     elif provider == ServiceProviders.OPENROUTER.value:
         kwargs = {}
         if base_url:
+            _validate_runtime_service_url(base_url, "base_url")
             kwargs["base_url"] = base_url
         return OpenRouterLLMService(
             api_key=api_key,
@@ -543,6 +565,8 @@ def create_llm_service_from_provider(
             settings=GoogleVertexLLMSettings(model=model, temperature=0.1),
         )
     elif provider == ServiceProviders.AZURE.value:
+        if endpoint:
+            _validate_runtime_service_url(endpoint, "endpoint")
         return AzureLLMService(
             api_key=api_key,
             endpoint=endpoint,
@@ -562,15 +586,19 @@ def create_llm_service_from_provider(
             settings=AWSBedrockLLMSettings(model=model),
         )
     elif provider == ServiceProviders.SPEACHES.value:
+        base_url = base_url or "http://localhost:11434/v1"
+        _validate_runtime_service_url(base_url, "base_url")
         return SpeachesLLMService(
-            base_url=base_url or "http://localhost:11434/v1",
+            base_url=base_url,
             api_key=api_key or "none",
             settings=SpeachesLLMSettings(model=model),
         )
     elif provider == ServiceProviders.MINIMAX.value:
+        base_url = base_url or "https://api.minimax.io/v1"
+        _validate_runtime_service_url(base_url, "base_url")
         return MiniMaxLLMService(
             api_key=api_key,
-            base_url=base_url or "https://api.minimax.io/v1",
+            base_url=base_url,
             settings=MiniMaxLLMService.Settings(
                 model=model,
                 temperature=temperature if temperature is not None else 1.0,
@@ -709,7 +737,9 @@ def create_llm_service(user_config):
     api_key = user_config.llm.api_key
 
     kwargs = {}
-    if provider == ServiceProviders.OPENROUTER.value:
+    if provider == ServiceProviders.OPENAI.value:
+        kwargs["base_url"] = user_config.llm.base_url
+    elif provider == ServiceProviders.OPENROUTER.value:
         kwargs["base_url"] = user_config.llm.base_url
     elif provider == ServiceProviders.AZURE.value:
         kwargs["endpoint"] = user_config.llm.endpoint

api/tests/test_user_configured_service_url_security.py

@@ -0,0 +1,227 @@
+from types import SimpleNamespace
+
+import pytest
+from fastapi import HTTPException
+
+from api.services.configuration.check_validity import UserConfigurationValidator
+from api.services.configuration.registry import (
+    ServiceProviders,
+    SpeachesLLMConfiguration,
+)
+from api.services.gen_ai.embedding.openai_service import OpenAIEmbeddingService
+from api.services.pipecat.service_factory import (
+    create_llm_service_from_provider,
+    create_tts_service,
+)
+from api.utils.url_security import validate_user_configured_service_url
+
+
+def test_oss_allows_local_service_urls(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "oss")
+
+    validate_user_configured_service_url(
+        "http://localhost:11434/v1",
+        field_name="base_url",
+    )
+
+
+@pytest.mark.parametrize(
+    "url",
+    [
+        "http://localhost:11434/v1",
+        "http://127.0.0.1:11434/v1",
+        "http://10.0.0.10/v1",
+        "http://169.254.169.254/latest/meta-data",
+        "http://100.64.0.1/v1",
+    ],
+)
+def test_saas_blocks_local_and_internal_service_urls(url, monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    with pytest.raises(ValueError):
+        validate_user_configured_service_url(
+            url,
+            field_name="base_url",
+        )
+
+
+def test_saas_rejects_unsupported_service_url_schemes(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    with pytest.raises(ValueError, match="http, https, ws, or wss"):
+        validate_user_configured_service_url(
+            "file:///etc/passwd",
+            field_name="base_url",
+        )
+
+
+def test_saas_checks_resolved_hostname_ips(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    def fake_getaddrinfo(*_args, **_kwargs):
+        return [(None, None, None, None, ("10.0.0.10", 443))]
+
+    monkeypatch.setattr("api.utils.url_security.socket.getaddrinfo", fake_getaddrinfo)
+
+    with pytest.raises(ValueError, match="public IP"):
+        validate_user_configured_service_url(
+            "https://internal.example.com/v1",
+            field_name="base_url",
+        )
+
+
+def test_saas_allows_public_service_url(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    def fake_getaddrinfo(*_args, **_kwargs):
+        return [(None, None, None, None, ("8.8.8.8", 443))]
+
+    monkeypatch.setattr("api.utils.url_security.socket.getaddrinfo", fake_getaddrinfo)
+
+    validate_user_configured_service_url(
+        "https://api.example.com/v1",
+        field_name="base_url",
+    )
+
+
+def test_saas_allows_public_websocket_service_url(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    def fake_getaddrinfo(*_args, **_kwargs):
+        return [(None, None, None, None, ("8.8.8.8", 443))]
+
+    monkeypatch.setattr("api.utils.url_security.socket.getaddrinfo", fake_getaddrinfo)
+
+    validate_user_configured_service_url(
+        "wss://api.example.com/v1",
+        field_name="base_url",
+    )
+
+
+def test_saas_blocks_local_websocket_service_url(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    with pytest.raises(ValueError, match="localhost"):
+        validate_user_configured_service_url(
+            "ws://localhost:8000/v1",
+            field_name="base_url",
+        )
+
+
+def test_validator_blocks_speaches_local_base_url_in_saas(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+    validator = UserConfigurationValidator()
+    config = SpeachesLLMConfiguration()
+
+    result = validator._validate_service(config, "llm")
+
+    assert result == [
+        {
+            "model": "llm",
+            "message": "base_url cannot point to localhost in SaaS mode",
+        }
+    ]
+
+
+def test_validator_blocks_azure_private_endpoint_in_saas(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+    validator = UserConfigurationValidator()
+    config = SimpleNamespace(
+        provider=ServiceProviders.AZURE.value,
+        endpoint="http://10.0.0.10/openai",
+        api_key="test-key",
+    )
+
+    result = validator._validate_service(config, "llm")
+
+    assert result == [
+        {
+            "model": "llm",
+            "message": "endpoint must resolve to a public IP address in SaaS mode",
+        }
+    ]
+
+
+def test_validator_allows_speaches_local_base_url_in_oss(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "oss")
+    validator = UserConfigurationValidator()
+    config = SpeachesLLMConfiguration()
+
+    assert validator._validate_service(config, "llm") == []
+
+
+def test_runtime_blocks_speaches_default_llm_base_url_in_saas(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    with pytest.raises(HTTPException) as exc_info:
+        create_llm_service_from_provider(
+            provider=ServiceProviders.SPEACHES.value,
+            model="llama3",
+            api_key=None,
+        )
+
+    assert exc_info.value.status_code == 400
+    assert "localhost" in exc_info.value.detail
+
+
+def test_runtime_blocks_openai_private_base_url_in_saas(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    with pytest.raises(HTTPException) as exc_info:
+        create_llm_service_from_provider(
+            provider=ServiceProviders.OPENAI.value,
+            model="gpt-4.1",
+            api_key="test-key",
+            base_url="http://10.0.0.10/v1",
+        )
+
+    assert exc_info.value.status_code == 400
+    assert "public IP" in exc_info.value.detail
+
+
+def test_runtime_blocks_azure_private_endpoint_in_saas(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+
+    with pytest.raises(HTTPException) as exc_info:
+        create_llm_service_from_provider(
+            provider=ServiceProviders.AZURE.value,
+            model="gpt-4.1-mini",
+            api_key="test-key",
+            endpoint="http://10.0.0.10/openai",
+        )
+
+    assert exc_info.value.status_code == 400
+    assert "public IP" in exc_info.value.detail
+
+
+def test_runtime_blocks_elevenlabs_local_tts_base_url_in_saas(monkeypatch):
+    monkeypatch.setattr("api.utils.url_security.DEPLOYMENT_MODE", "saas")
+    user_config = SimpleNamespace(
+        tts=SimpleNamespace(
+            provider=ServiceProviders.ELEVENLABS.value,
+            api_key="test-key",
+            model="eleven_flash_v2_5",
+            voice="voice-id",
+            speed=1.0,
+            base_url="http://localhost:8000",
+        )
+    )
+
+    with pytest.raises(HTTPException) as exc_info:
+        create_tts_service(user_config, audio_config=None)
+
+    assert exc_info.value.status_code == 400
+    assert "localhost" in exc_info.value.detail
+
+
+def test_embedding_service_blocks_private_base_url_in_saas(monkeypatch):
+    monkeypatch.setattr(
+        "api.utils.url_security.DEPLOYMENT_MODE", "saas"
+    )
+
+    with pytest.raises(ValueError, match="public IP"):
+        OpenAIEmbeddingService(
+            db_client=SimpleNamespace(),
+            api_key="test-key",
+            base_url="http://10.0.0.10/v1",
+        )

api/utils/url_security.py

@@ -0,0 +1,66 @@
+import ipaddress
+import socket
+from urllib.parse import urlparse
+
+from api.constants import DEPLOYMENT_MODE
+
+_CGNAT_NETWORK = ipaddress.ip_network("100.64.0.0/10")
+
+
+def validate_user_configured_service_url(
+    url: str,
+    *,
+    field_name: str,
+) -> None:
+    """Restrict user-configured service URLs in hosted deployments.
+
+    OSS deployments commonly point model services at localhost or private LAN
+    hosts. SaaS deployments must not allow users to make Dograh infrastructure
+    connect to private/internal network locations.
+    """
+    if DEPLOYMENT_MODE == "oss":
+        return
+
+    parsed = urlparse(url)
+    if parsed.scheme not in {"http", "https", "ws", "wss"} or not parsed.hostname:
+        raise ValueError(f"{field_name} must be an http, https, ws, or wss URL")
+
+    hostname = parsed.hostname
+    if hostname.lower() == "localhost":
+        raise ValueError(f"{field_name} cannot point to localhost in SaaS mode")
+
+    for ip in _resolve_hostname_ips(hostname, parsed.port):
+        if _is_blocked_saas_service_ip(ip):
+            raise ValueError(
+                f"{field_name} must resolve to a public IP address in SaaS mode"
+            )
+
+
+def _resolve_hostname_ips(
+    hostname: str, port: int | None
+) -> list[ipaddress.IPv4Address | ipaddress.IPv6Address]:
+    try:
+        return [ipaddress.ip_address(hostname)]
+    except ValueError:
+        pass
+
+    try:
+        addr_infos = socket.getaddrinfo(hostname, port, type=socket.SOCK_STREAM)
+    except socket.gaierror as e:
+        raise ValueError("Could not resolve service URL hostname") from e
+
+    return [ipaddress.ip_address(addr_info[4][0]) for addr_info in addr_infos]
+
+
+def _is_blocked_saas_service_ip(
+    ip: ipaddress.IPv4Address | ipaddress.IPv6Address,
+) -> bool:
+    return (
+        ip.is_private
+        or ip.is_loopback
+        or ip.is_link_local
+        or ip.is_multicast
+        or ip.is_reserved
+        or ip.is_unspecified
+        or (ip.version == 4 and ip in _CGNAT_NETWORK)
+    )