Custom OpenAI api base

docker-compose.yml

@@ -68,6 +68,7 @@ services:
       - OPENSEARCH_USERNAME=${OPENSEARCH_USERNAME:-admin}
       - OPENSEARCH_PASSWORD=${OPENSEARCH_PASSWORD}
       - OPENAI_API_KEY=${OPENAI_API_KEY}
+      - OPENAI_API_BASE=${OPENAI_API_BASE}
       - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
       - WATSONX_API_KEY=${WATSONX_API_KEY}
       - WATSONX_ENDPOINT=${WATSONX_ENDPOINT}
@@ -132,6 +133,7 @@ services:
       - LANGFUSE_HOST=${LANGFUSE_HOST:-}
       - LANGFLOW_DEACTIVATE_TRACING
       - OPENAI_API_KEY=${OPENAI_API_KEY:-None}
+      - OPENAI_API_BASE=${OPENAI_API_BASE:-None}
       - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY:-None}
       - WATSONX_API_KEY=${WATSONX_API_KEY:-None}
       - WATSONX_ENDPOINT=${WATSONX_ENDPOINT:-None}
@@ -160,7 +162,7 @@ services:
       - MIMETYPE=None
       - FILESIZE=0
       - SELECTED_EMBEDDING_MODEL=${SELECTED_EMBEDDING_MODEL:-}
-      - LANGFLOW_VARIABLES_TO_GET_FROM_ENVIRONMENT=JWT,OPENRAG-QUERY-FILTER,OPENSEARCH_PASSWORD,OPENSEARCH_URL,DOCLING_SERVE_URL,OWNER,OWNER_NAME,OWNER_EMAIL,CONNECTOR_TYPE,DOCUMENT_ID,SOURCE_URL,ALLOWED_USERS,ALLOWED_GROUPS,FILENAME,MIMETYPE,FILESIZE,SELECTED_EMBEDDING_MODEL,OPENAI_API_KEY,ANTHROPIC_API_KEY,WATSONX_API_KEY,WATSONX_ENDPOINT,WATSONX_PROJECT_ID,OLLAMA_BASE_URL,OPENSEARCH_INDEX_NAME
+      - LANGFLOW_VARIABLES_TO_GET_FROM_ENVIRONMENT=JWT,OPENRAG-QUERY-FILTER,OPENSEARCH_PASSWORD,OPENSEARCH_URL,DOCLING_SERVE_URL,OWNER,OWNER_NAME,OWNER_EMAIL,CONNECTOR_TYPE,DOCUMENT_ID,SOURCE_URL,ALLOWED_USERS,ALLOWED_GROUPS,FILENAME,MIMETYPE,FILESIZE,SELECTED_EMBEDDING_MODEL,OPENAI_API_KEY,OPENAI_API_BASE,ANTHROPIC_API_KEY,WATSONX_API_KEY,WATSONX_ENDPOINT,WATSONX_PROJECT_ID,OLLAMA_BASE_URL,OPENSEARCH_INDEX_NAME
       - LANGFLOW_LOG_LEVEL=DEBUG
       - LANGFLOW_WORKERS=${LANGFLOW_WORKERS:-1}
       - LANGFLOW_AUTO_LOGIN=${LANGFLOW_AUTO_LOGIN}

flows/ingestion_flow.json

@@ -5565,7 +5565,7 @@
               "api_base": {
                 "_input_type": "MessageTextInput",
                 "advanced": true,
-                "display_name": "OpenAI API Base URL",
+                "display_name": "API Base URL",
                 "dynamic": false,
                 "info": "Base URL for the API. Leave empty for default.",
                 "input_types": [

flows/openrag_agent.json

@@ -1806,28 +1806,6 @@
                 "type": "int",
                 "value": 100
               },
-              "openai_api_base": {
-                "_input_type": "StrInput",
-                "advanced": true,
-                "display_name": "OpenAI API Base",
-                "dynamic": false,
-                "info": "The base URL of the OpenAI API. Defaults to https://api.openai.com/v1. You can change this to use other APIs like JinaChat, LocalAI and Prem.",
-                "input_types": [],
-                "list": false,
-                "list_add_label": "Add More",
-                "load_from_db": false,
-                "name": "openai_api_base",
-                "override_skip": false,
-                "placeholder": "",
-                "required": false,
-                "show": true,
-                "title_case": false,
-                "tool_mode": false,
-                "trace_as_metadata": true,
-                "track_in_telemetry": false,
-                "type": "str",
-                "value": ""
-              },
               "output_schema": {
                 "_input_type": "TableInput",
                 "advanced": true,

src/api/models.py

@@ -1,3 +1,4 @@
+import os
 from typing import Optional
 
 from fastapi import Depends
@@ -45,8 +46,20 @@ async def get_openai_models(
                 {"error": "OpenAI API key is required either in request body or in configuration"},
                 status_code=400,
             )
+
+        api_base = None
+        try:
+            config = get_openrag_config()
+            api_base = config.providers.openai.endpoint
+            logger.info(
+                f"Retrieved OpenAI API base from config: {'yes' if api_base else 'no'}"
+            )
+        except Exception as e:
+            logger.error(f"Failed to get config: {e}")
+        if not api_base:
+            api_base = os.environ.get("OPENAI_API_BASE", "https://api.openai.com")
 
-        models = await models_service.get_openai_models(api_key=api_key)
+        models = await models_service.get_openai_models(api_key=api_key, api_base=api_base)
         return JSONResponse(models)
     except Exception as e:
         logger.error(f"Failed to get OpenAI models: {str(e)}")

src/api/provider_validation.py

@@ -1,6 +1,8 @@
 """Provider validation utilities for testing API keys and models during onboarding."""
 
 import json
+import os
+
 import httpx
 from utils.container_utils import transform_localhost_url
 from utils.logging_config import get_logger
@@ -105,7 +107,6 @@ def _extract_error_details(response: httpx.Response) -> str:
             return parsed
         return response_text
 
-
 async def validate_provider_setup(
     provider: str,
     api_key: str = None,
@@ -123,7 +124,7 @@ async def validate_provider_setup(
         api_key: API key for the provider (optional for ollama)
         embedding_model: Embedding model to test
         llm_model: LLM model to test
-        endpoint: Provider endpoint (required for ollama and watsonx)
+        endpoint: Provider endpoint (required for ollama and watsonx, optional for openai)
         project_id: Project ID (required for watsonx)
         test_completion: If True, performs full validation with completion/embedding tests (consumes credits).
                         If False, performs lightweight validation (no credits consumed). Default: False.
@@ -136,11 +137,19 @@ async def validate_provider_setup(
     try:
         logger.info(f"Starting validation for provider: {provider_lower} (test_completion={test_completion})")
 
+        if provider == "openai" and not endpoint:
+            endpoint = os.environ.get("OPENAI_API_BASE", "https://api.openai.com")
+
+        # Strip /v1 suffix from OpenAI endpoint if present to avoid double /v1 paths
+        if provider == "openai" and endpoint and endpoint.endswith("/v1"):
+            endpoint = endpoint.rstrip("/v1")
+            logger.info(f"Stripped /v1 suffix from OpenAI endpoint: {endpoint}")
+
         if test_completion:
             # Full validation with completion/embedding tests (consumes credits)
             if embedding_model:
                 # Test embedding
-                await test_embedding(
+                await _test_embedding(
                     provider=provider_lower,
                     api_key=api_key,
                     embedding_model=embedding_model,
@@ -149,7 +158,7 @@ async def validate_provider_setup(
                 )
             elif llm_model:
                 # Test completion with tool calling
-                await test_completion_with_tools(
+                await _test_completion_with_tools(
                     provider=provider_lower,
                     api_key=api_key,
                     llm_model=llm_model,
@@ -158,7 +167,7 @@ async def validate_provider_setup(
                 )
         else:
             # Lightweight validation (no credits consumed)
-            await test_lightweight_health(
+            await _test_lightweight_health(
                 provider=provider_lower,
                 api_key=api_key,
                 endpoint=endpoint,
@@ -173,7 +182,7 @@ async def validate_provider_setup(
         raise
 
 
-async def test_lightweight_health(
+async def _test_lightweight_health(
     provider: str,
     api_key: str = None,
     endpoint: str = None,
@@ -182,7 +191,7 @@ async def test_lightweight_health(
     """Test provider health with lightweight check (no credits consumed)."""
 
     if provider == "openai":
-        await _test_openai_lightweight_health(api_key)
+        await _test_openai_lightweight_health(api_key, endpoint)
     elif provider == "watsonx":
         await _test_watsonx_lightweight_health(api_key, endpoint, project_id)
     elif provider == "ollama":
@@ -193,7 +202,7 @@ async def test_lightweight_health(
         raise ValueError(f"Unknown provider: {provider}")
 
 
-async def test_completion_with_tools(
+async def _test_completion_with_tools(
     provider: str,
     api_key: str = None,
     llm_model: str = None,
@@ -203,7 +212,7 @@ async def test_completion_with_tools(
     """Test completion with tool calling for the provider."""
 
     if provider == "openai":
-        await _test_openai_completion_with_tools(api_key, llm_model)
+        await _test_openai_completion_with_tools(api_key, llm_model, endpoint)
     elif provider == "watsonx":
         await _test_watsonx_completion_with_tools(api_key, llm_model, endpoint, project_id)
     elif provider == "ollama":
@@ -214,7 +223,7 @@ async def test_completion_with_tools(
         raise ValueError(f"Unknown provider: {provider}")
 
 
-async def test_embedding(
+async def _test_embedding(
     provider: str,
     api_key: str = None,
     embedding_model: str = None,
@@ -224,7 +233,7 @@ async def test_embedding(
     """Test embedding generation for the provider."""
 
     if provider == "openai":
-        await _test_openai_embedding(api_key, embedding_model)
+        await _test_openai_embedding(api_key, embedding_model, endpoint)
     elif provider == "watsonx":
         await _test_watsonx_embedding(api_key, embedding_model, endpoint, project_id)
     elif provider == "ollama":
@@ -234,7 +243,7 @@ async def test_embedding(
 
 
 # OpenAI validation functions
-async def _test_openai_lightweight_health(api_key: str) -> None:
+async def _test_openai_lightweight_health(api_key: str, endpoint: str) -> None:
     """Test OpenAI API key validity with lightweight check.
 
     Only checks if the API key is valid without consuming credits.
@@ -246,10 +255,12 @@ async def _test_openai_lightweight_health(api_key: str) -> None:
             "Content-Type": "application/json",
         }
 
+        url = f"{endpoint}/v1/models"
+        logger.info("Testing openai lightweight health", url=url)
         async with httpx.AsyncClient() as client:
             # Use /v1/models endpoint which validates the key without consuming credits
             response = await client.get(
-                "https://api.openai.com/v1/models",
+                url=url,
                 headers=headers,
                 timeout=10.0,  # Short timeout for lightweight check
             )
@@ -269,7 +280,7 @@ async def _test_openai_lightweight_health(api_key: str) -> None:
         raise
 
 
-async def _test_openai_completion_with_tools(api_key: str, llm_model: str) -> None:
+async def _test_openai_completion_with_tools(api_key: str, llm_model: str, endpoint: str) -> None:
     """Test OpenAI completion with tool calling."""
     try:
         headers = {
@@ -307,8 +318,10 @@ async def _test_openai_completion_with_tools(api_key: str, llm_model: str) -> No
         async with httpx.AsyncClient() as client:
             # Try with max_tokens first
             payload = {**base_payload, "max_tokens": 50}
+            url = f"{endpoint}/v1/chat/completions"
+            logger.info("Test openai completion tools", url=url)
             response = await client.post(
-                "https://api.openai.com/v1/chat/completions",
+                url=url,
                 headers=headers,
                 json=payload,
                 timeout=30.0,
@@ -318,8 +331,9 @@ async def _test_openai_completion_with_tools(api_key: str, llm_model: str) -> No
             if response.status_code != 200:
                 logger.info("max_tokens parameter failed, trying max_completion_tokens instead")
                 payload = {**base_payload, "max_completion_tokens": 50}
+                logger.info("Test openai completion tools", url=url)
                 response = await client.post(
-                    "https://api.openai.com/v1/chat/completions",
+                    url=url,
                     headers=headers,
                     json=payload,
                     timeout=30.0,
@@ -340,7 +354,7 @@ async def _test_openai_completion_with_tools(api_key: str, llm_model: str) -> No
         raise
 
 
-async def _test_openai_embedding(api_key: str, embedding_model: str) -> None:
+async def _test_openai_embedding(api_key: str, embedding_model: str, endpoint: str) -> None:
     """Test OpenAI embedding generation."""
     try:
         headers = {
@@ -354,8 +368,9 @@ async def _test_openai_embedding(api_key: str, embedding_model: str) -> None:
         }
 
         async with httpx.AsyncClient() as client:
+            url = f"{endpoint}/v1/embeddings"
             response = await client.post(
-                "https://api.openai.com/v1/embeddings",
+                url=url,
                 headers=headers,
                 json=payload,
                 timeout=30.0,

src/config/config_manager.py

@@ -14,6 +14,7 @@
 class OpenAIConfig:
     """OpenAI provider configuration."""
     api_key: str = ""
+    endpoint: str = ""
     configured: bool = False
 
 
@@ -224,6 +225,8 @@ def _load_env_overrides(
         # OpenAI provider settings
         if os.getenv("OPENAI_API_KEY"):
             config_data["providers"]["openai"]["api_key"] = os.getenv("OPENAI_API_KEY")
+        if os.getenv("OPENAI_API_BASE"):
+            config_data["providers"]["openai"]["endpoint"] = os.getenv("OPENAI_API_BASE")
 
         # Anthropic provider settings
         if os.getenv("ANTHROPIC_API_KEY"):

src/config/settings.py

@@ -439,7 +439,10 @@ def patched_async_client(self):
                 if config.providers.openai.api_key:
                     os.environ["OPENAI_API_KEY"] = config.providers.openai.api_key
                     logger.debug("Loaded OpenAI API key from config")
-
+                if config.providers.openai.endpoint:
+                    os.environ["OPENAI_API_BASE"] = config.providers.openai.endpoint
+                    logger.debug(f"Loaded OpenAI endpoint from config: '{config.providers.openai.endpoint}'")
+
                 # Set Anthropic credentials
                 if config.providers.anthropic.api_key:
                     os.environ["ANTHROPIC_API_KEY"] = config.providers.anthropic.api_key

src/services/flows_service.py

@@ -1393,11 +1393,17 @@ async def _update_component_fields(
             template["api_key"]["advanced"] = False
             updated = True
         if provider == "openai" and "api_base" in template:
-            template["api_base"]["value"] = ""
-            template["api_base"]["load_from_db"] = False
+            template["api_base"]["value"] = "OPENAI_API_BASE"
+            template["api_base"]["load_from_db"] = True
             template["api_base"]["show"] = True
             template["api_base"]["advanced"] = False
             updated = True
+        if provider == "openai" and "openai_api_base" in template:
+            template["openai_api_base"]["value"] = "OPENAI_API_BASE"
+            template["openai_api_base"]["load_from_db"] = True
+            template["openai_api_base"]["show"] = True
+            template["openai_api_base"]["advanced"] = False
+            updated = True
 
         if provider == "anthropic" and "api_key" in template:
             template["api_key"]["value"] = "ANTHROPIC_API_KEY"

src/services/models_service.py

@@ -19,7 +19,7 @@ class ModelsService:
     def __init__(self):
         self.session_manager = None
 
-    async def get_openai_models(self, api_key: str) -> Dict[str, List[Dict[str, str]]]:
+    async def get_openai_models(self, api_key: str, api_base: str) -> Dict[str, List[Dict[str, str]]]:
         """Fetch available models from OpenAI API with lightweight validation"""
         try:
             headers = {
@@ -30,8 +30,10 @@ async def get_openai_models(self, api_key: str) -> Dict[str, List[Dict[str, str]
             async with httpx.AsyncClient() as client:
                 # Lightweight validation: just check if API key is valid
                 # This doesn't consume credits, only validates the key
+                url = f"{api_base}/v1/models"
+                logger.debug("Getting openai models.", url=url)
                 response = await client.get(
-                    "https://api.openai.com/v1/models", headers=headers, timeout=10.0
+                    url, headers=headers, timeout=10.0
                 )
 
             if response.status_code == 200:

src/utils/langflow_headers.py

@@ -14,7 +14,10 @@ def add_provider_credentials_to_headers(headers: Dict[str, str], config) -> None
     # Add OpenAI credentials
     if config.providers.openai.api_key:
         headers["X-LANGFLOW-GLOBAL-VAR-OPENAI_API_KEY"] = str(config.providers.openai.api_key)
-
+
+    if config.providers.openai.endpoint:
+        headers["X-LANGFLOW-GLOBAL-VAR-OPENAI_API_BASE"] = str(config.providers.openai.endpoint)
+
     # Add Anthropic credentials
     if config.providers.anthropic.api_key:
         headers["X-LANGFLOW-GLOBAL-VAR-ANTHROPIC_API_KEY"] = str(config.providers.anthropic.api_key)
@@ -47,6 +50,9 @@ def build_mcp_global_vars_from_config(config) -> Dict[str, str]:
     if config.providers.openai.api_key:
         global_vars["OPENAI_API_KEY"] = config.providers.openai.api_key
 
+    if config.providers.openai.endpoint:
+        global_vars["OPENAI_API_BASE"] = config.providers.openai.endpoint
+
     # Add Anthropic credentials
     if config.providers.anthropic.api_key:
         global_vars["ANTHROPIC_API_KEY"] = config.providers.anthropic.api_key