feat(llm): enhance LiteLLMCaller to support custom API key handling a…

backend/modules/llm/litellm_caller.py

@@ -85,11 +85,13 @@ def _get_model_kwargs(self, model_name: str, temperature: Optional[float] = None
         except ValueError as e:
             logger.error(f"Failed to resolve API key for model {model_name}: {e}")
             raise
-        
+
         if api_key:
+            # Always pass api_key to LiteLLM for all providers
+            kwargs["api_key"] = api_key
+
+            # Additionally set provider-specific env vars for LiteLLM's internal logic
             if "openrouter" in model_config.model_url:
-                kwargs["api_key"] = api_key
-                # LiteLLM will automatically set the correct env var
                 os.environ["OPENROUTER_API_KEY"] = api_key
             elif "openai" in model_config.model_url:
                 os.environ["OPENAI_API_KEY"] = api_key
@@ -99,6 +101,10 @@ def _get_model_kwargs(self, model_name: str, temperature: Optional[float] = None
                 os.environ["GOOGLE_API_KEY"] = api_key
             elif "cerebras" in model_config.model_url:
                 os.environ["CEREBRAS_API_KEY"] = api_key
+            else:
+                # Custom endpoint - set OPENAI_API_KEY as fallback
+                # (most custom endpoints are OpenAI-compatible)
+                os.environ["OPENAI_API_KEY"] = api_key
 
         # Set custom API base for non-standard endpoints
         if hasattr(model_config, 'model_url') and model_config.model_url:

backend/tests/test_llm_env_expansion.py

@@ -146,14 +146,148 @@ def test_litellm_caller_handles_literal_extra_headers(self):
                 )
             }
         )
-        
+
         # Create LiteLLMCaller
         caller = LiteLLMCaller(llm_config, debug_mode=True)
-        
+
         # Get model kwargs - this should work without errors
         model_kwargs = caller._get_model_kwargs("test-model")
-        
+
         # Verify that extra_headers were passed through
         assert "extra_headers" in model_kwargs
         assert model_kwargs["extra_headers"]["HTTP-Referer"] == "https://literal-app.com"
         assert model_kwargs["extra_headers"]["X-Title"] == "LiteralApp"
+
+    def test_custom_endpoint_with_env_var_api_key(self, monkeypatch):
+        """Custom endpoint should pass api_key in kwargs when using env var."""
+        monkeypatch.setenv("CUSTOM_LLM_KEY", "sk-custom-12345")
+
+        # Create LLM config for custom endpoint with env var in api_key
+        llm_config = LLMConfig(
+            models={
+                "custom-model": ModelConfig(
+                    model_name="custom-model-name",
+                    model_url="https://custom-llm.example.com/v1",
+                    api_key="${CUSTOM_LLM_KEY}"
+                )
+            }
+        )
+
+        # Create LiteLLMCaller
+        caller = LiteLLMCaller(llm_config, debug_mode=True)
+
+        # Get model kwargs
+        model_kwargs = caller._get_model_kwargs("custom-model")
+
+        # Verify that api_key is in kwargs (critical for custom endpoints)
+        assert "api_key" in model_kwargs
+        assert model_kwargs["api_key"] == "sk-custom-12345"
+
+        # Verify that api_base is set for custom endpoint
+        assert "api_base" in model_kwargs
+        assert model_kwargs["api_base"] == "https://custom-llm.example.com/v1"
+
+        # Verify fallback env var is set for OpenAI-compatible endpoints
+        import os
+        assert os.environ.get("OPENAI_API_KEY") == "sk-custom-12345"
+
+    def test_custom_endpoint_with_literal_api_key(self):
+        """Custom endpoint should pass api_key in kwargs when using literal value."""
+        # Create LLM config for custom endpoint with literal api_key
+        llm_config = LLMConfig(
+            models={
+                "custom-model": ModelConfig(
+                    model_name="custom-model-name",
+                    model_url="https://custom-llm.example.com/v1",
+                    api_key="sk-literal-custom-key"
+                )
+            }
+        )
+
+        # Create LiteLLMCaller
+        caller = LiteLLMCaller(llm_config, debug_mode=True)
+
+        # Get model kwargs
+        model_kwargs = caller._get_model_kwargs("custom-model")
+
+        # Verify that api_key is in kwargs (critical for custom endpoints)
+        assert "api_key" in model_kwargs
+        assert model_kwargs["api_key"] == "sk-literal-custom-key"
+
+        # Verify that api_base is set for custom endpoint
+        assert "api_base" in model_kwargs
+        assert model_kwargs["api_base"] == "https://custom-llm.example.com/v1"
+
+    def test_custom_endpoint_with_extra_headers(self, monkeypatch):
+        """Custom endpoint should handle extra_headers correctly."""
+        monkeypatch.setenv("CUSTOM_API_KEY", "sk-custom-auth")
+        monkeypatch.setenv("CUSTOM_TENANT", "tenant-123")
+
+        # Create LLM config for custom endpoint with extra headers
+        llm_config = LLMConfig(
+            models={
+                "custom-model": ModelConfig(
+                    model_name="custom-model-name",
+                    model_url="https://custom-llm.example.com/v1",
+                    api_key="${CUSTOM_API_KEY}",
+                    extra_headers={
+                        "X-Tenant-ID": "${CUSTOM_TENANT}",
+                        "X-Custom-Header": "custom-value"
+                    }
+                )
+            }
+        )
+
+        # Create LiteLLMCaller
+        caller = LiteLLMCaller(llm_config, debug_mode=True)
+
+        # Get model kwargs
+        model_kwargs = caller._get_model_kwargs("custom-model")
+
+        # Verify api_key is passed
+        assert "api_key" in model_kwargs
+        assert model_kwargs["api_key"] == "sk-custom-auth"
+
+        # Verify extra_headers are resolved and passed
+        assert "extra_headers" in model_kwargs
+        assert model_kwargs["extra_headers"]["X-Tenant-ID"] == "tenant-123"
+        assert model_kwargs["extra_headers"]["X-Custom-Header"] == "custom-value"
+
+        # Verify api_base is set
+        assert "api_base" in model_kwargs
+
+    def test_known_providers_still_get_api_key_in_kwargs(self):
+        """Verify that known providers also get api_key in kwargs (backward compatibility)."""
+        # Test OpenAI
+        llm_config = LLMConfig(
+            models={
+                "openai-model": ModelConfig(
+                    model_name="gpt-4",
+                    model_url="https://api.openai.com/v1",
+                    api_key="sk-openai-test"
+                )
+            }
+        )
+        caller = LiteLLMCaller(llm_config, debug_mode=True)
+        model_kwargs = caller._get_model_kwargs("openai-model")
+
+        # OpenAI should get api_key in kwargs
+        assert "api_key" in model_kwargs
+        assert model_kwargs["api_key"] == "sk-openai-test"
+
+        # Test OpenRouter
+        llm_config = LLMConfig(
+            models={
+                "openrouter-model": ModelConfig(
+                    model_name="meta-llama/llama-3-70b",
+                    model_url="https://openrouter.ai/api/v1",
+                    api_key="sk-or-test"
+                )
+            }
+        )
+        caller = LiteLLMCaller(llm_config, debug_mode=True)
+        model_kwargs = caller._get_model_kwargs("openrouter-model")
+
+        # OpenRouter should get api_key in kwargs
+        assert "api_key" in model_kwargs
+        assert model_kwargs["api_key"] == "sk-or-test"