diff --git a/apps/knowledge/tasks.py b/apps/knowledge/tasks.py
index ffac9fe..3889724 100644
--- a/apps/knowledge/tasks.py
+++ b/apps/knowledge/tasks.py
@@ -9,6 +9,7 @@ from pypdf import PdfReader
 
 from apps.knowledge.models import RoleRagDocument, TrainingFile
 
+
 def _decode_text_bytes(raw_bytes: bytes) -> str:
     try:
         return raw_bytes.decode('utf-8')
@@ -63,7 +64,7 @@ def ingest_training_file_task(self, file_uuid):
                                                               
             for text_segment in _get_text_chunks(raw_text):
                 response = client.post(
-                    f"{settings.INFERENCE_URL}/v1/semantic-chunk",
+                    settings.INFERENCE_SEMANTIC_CHUNK_ENDPOINT,
                     json={"text": text_segment, "threshold": 95}
                 )
                 response.raise_for_status()
diff --git a/apps/onboarding/consumers.py b/apps/onboarding/consumers.py
index ec5f49b..53d8955 100644
--- a/apps/onboarding/consumers.py
+++ b/apps/onboarding/consumers.py
@@ -305,7 +305,7 @@ class OnboardingConsumer(AsyncWebsocketConsumer):
                 
                 try:
                     response = await client.post(
-                        f"{settings.INFERENCE_URL}/v1/chat/completions",
+                        settings.INFERENCE_CHAT_COMPLETIONS_ENDPOINT,
                         json={
                             "model": llm_config.get("model_id", "meta-llama-3.1-8b"),
                             "messages": messages,
diff --git a/apps/onboarding/mcp.py b/apps/onboarding/mcp.py
index 38a1220..d04e384 100644
--- a/apps/onboarding/mcp.py
+++ b/apps/onboarding/mcp.py
@@ -6,6 +6,7 @@ from pgvector.django import CosineDistance
 from apps.knowledge.models import RoleRagDocument
 from apps.onboarding.models import OnboardingSession
 
+
 class MCPRouter:
 
     def get_tool_definitions(self):
@@ -47,7 +48,7 @@ class MCPRouter:
     async def _get_embedding(self, text):
         async with httpx.AsyncClient() as client:
             response = await client.post(
-                f"{settings.INFERENCE_URL}/v1/embeddings",                                           
+                settings.INFERENCE_EMBEDDINGS_ENDPOINT,
                 json={"input": text}
             )
                                                          
diff --git a/apps/onboarding/viewsets.py b/apps/onboarding/viewsets.py
index b380551..2f38a06 100644
--- a/apps/onboarding/viewsets.py
+++ b/apps/onboarding/viewsets.py
@@ -15,6 +15,7 @@ from apps.accounts.permissions import CanManageOrganization, can_manage_organiza
 from apps.onboarding.models import AgentConfig, AgentInteractionLog, OnboardingFlow, OnboardingSession
 from apps.onboarding.serializers import AgentConfigSerializer, AgentInteractionLogSerializer, OnboardingFlowSerializer, OnboardingSessionSerializer
 
+
 class OnboardingFlowViewSet(ModelViewSet):
     queryset = OnboardingFlow.objects.all()
     serializer_class = OnboardingFlowSerializer
@@ -445,7 +446,7 @@ class OnboardingSessionViewSet(ModelViewSet):
         try:
             with httpx.Client(timeout=60.0) as client:
                 response = client.post(
-                    f"{settings.INFERENCE_URL}/v1/chat/completions",
+                    settings.INFERENCE_CHAT_COMPLETIONS_ENDPOINT,
                     json={
                         "model": (config.llm_config or {}).get("model_id", "meta-llama-3.1-8b"),
                         "messages": [
@@ -482,7 +483,7 @@ class OnboardingSessionViewSet(ModelViewSet):
         try:
             with httpx.Client(timeout=60.0) as client:
                 response = client.post(
-                    f"{settings.INFERENCE_URL}/v1/chat/completions",
+                    settings.INFERENCE_CHAT_COMPLETIONS_ENDPOINT,
                     json={
                         "model": (config.llm_config or {}).get("model_id", "meta-llama-3.1-8b"),
                         "messages": [
diff --git a/config/settings.py b/config/settings.py
index f027eb9..0a16cf7 100644
--- a/config/settings.py
+++ b/config/settings.py
@@ -3,8 +3,9 @@ Django settings will use prefix of DJANGO_ for environment variables.
 """
 
 import os
-from pathlib import Path
 import sys
+from pathlib import Path
+
 from dotenv import load_dotenv
 
 BASE_DIR = Path(__file__).resolve().parent.parent
@@ -27,6 +28,9 @@ DJANGO_CELERY_BROKER_URL = os.getenv('DJANGO_CELERY_BROKER_URL', 'redis://localh
 INFERENCE_HOST = os.getenv('INFERENCE_HOST', 'localhost')
 INFERENCE_PORT = os.getenv('INFERENCE_PORT', '8001')
 INFERENCE_URL = f"http://{INFERENCE_HOST}:{INFERENCE_PORT}"
+INFERENCE_SEMANTIC_CHUNK_ENDPOINT = f"{INFERENCE_URL}/v1/semantic-chunk"
+INFERENCE_EMBEDDINGS_ENDPOINT = f"{INFERENCE_URL}/v1/embeddings"
+INFERENCE_CHAT_COMPLETIONS_ENDPOINT = f"{INFERENCE_URL}/v1/chat/completions"
 INFERENCE_INGEST_TIMEOUT = float(os.getenv('INFERENCE_INGEST_TIMEOUT', '600'))
 
 STATIC_URL  = os.getenv('DJANGO_STATIC_URL',  '/static/')