diff --git a/backend/.env.example b/backend/.env.example
index e7a3cba..e22ddba 100644
--- a/backend/.env.example
+++ b/backend/.env.example
@@ -21,6 +21,9 @@ OPENCAD_AGENT_LIVE_KERNEL=false
 OPENCAD_TREE_LIVE_KERNEL=true
 OPENCAD_AGENT_LIVE_KERNEL=true
 
+OPENCAD_LLM_PROVIDER=
+OPENCAD_LLM_MODEL=
+
 # Optional multi-provider LLM support for code generation uses LiteLLM.
 # Provider-specific credentials are still read from each provider's standard env vars.
 # Example request fields: llm_provider="openai", llm_model="gpt-4o-mini", generate_code=true
diff --git a/backend/opencad_agent/llm.py b/backend/opencad_agent/llm.py
index 4b74d72..cf17d17 100644
--- a/backend/opencad_agent/llm.py
+++ b/backend/opencad_agent/llm.py
@@ -20,12 +20,6 @@ def _default_completion(**kwargs: Any) -> Any:
     return completion(**kwargs)
 
 
-def _resolve_model_name(provider: str | None, model: str) -> str:
-    if provider and "/" not in model:
-        return f"{provider}/{model}"
-    return model
-
-
 def _strip_code_fences(code: str) -> str:
     code = code.strip()
     if code.startswith("```python"):
@@ -68,7 +62,6 @@ def __init__(self, completion_func: LiteLlmCompletion | None = None) -> None:
     def generate_code(
         self,
         *,
-        provider: str | None,
         model: str,
         system_prompt: str,
         user_message: str,
@@ -79,9 +72,10 @@ def generate_code(
         messages.extend({"role": item.role, "content": item.content} for item in conversation_history)
         messages.append({"role": "user", "content": user_message})
         response = self._completion(
-            model=_resolve_model_name(provider, model),
+            model=model,
             messages=messages,
             temperature=HIGH_REASONING_CODE_TEMPERATURE if reasoning else DEFAULT_CODE_TEMPERATURE,
         )
         logger.debug("Received LLM code-generation response")
-        return _strip_code_fences(_extract_message_content(response))
+        cleaned_response = _strip_code_fences(_extract_message_content(response))
+        return cleaned_response
diff --git a/backend/opencad_agent/models.py b/backend/opencad_agent/models.py
index c3720b7..b6de661 100644
--- a/backend/opencad_agent/models.py
+++ b/backend/opencad_agent/models.py
@@ -17,15 +17,8 @@ class ChatRequest(BaseModel):
     tree_state: FeatureTree
     conversation_history: list[ChatHistoryItem] = Field(default_factory=list)
     reasoning: bool = False
-    llm_provider: str | None = None
-    llm_model: str | None = None
     generate_code: bool = False
 
-    @model_validator(mode="after")
-    def _validate_llm_configuration(self) -> ChatRequest:
-        if self.llm_provider and not self.llm_model:
-            raise ValueError("llm_model is required when llm_provider is set.")
-        return self
 
 
 class OperationExecution(BaseModel):
diff --git a/backend/opencad_agent/service.py b/backend/opencad_agent/service.py
index c29a802..7971498 100644
--- a/backend/opencad_agent/service.py
+++ b/backend/opencad_agent/service.py
@@ -124,11 +124,10 @@ def _execute_code_in_context(code: str, ctx: object) -> None:
             reset_default_context()
 
     def _generate_code(self, request: ChatRequest) -> str:
-        provider = request.llm_provider or os.environ.get("OPENCAD_LLM_PROVIDER")
-        model = request.llm_model or os.environ.get("OPENCAD_LLM_MODEL")
+        model = os.environ.get("OPENCAD_LLM_MODEL")
+
         if model:
             return self.llm_client.generate_code(
-                provider=provider,
                 model=model,
                 system_prompt=build_code_generation_prompt(request.tree_state),
                 user_message=request.message,