feat: Modify endpoints for OpenAPI compatibility

docs/user-guides/community/openai.md

@@ -0,0 +1,16 @@
+## OpenAI API Compatibility for NeMo Guardrails
+
+NeMo Guardrails provides server-side compatibility with OpenAI API endpoints, enabling applications that use OpenAI clients to seamlessly integrate with NeMo Guardrails for adding guardrails to LLM interactions. Point your OpenAI client to `http://localhost:8000` (or your server URL) and use the standard `/v1/chat/completions` endpoint.
+
+## Feature Support Matrix
+
+The following table outlines which OpenAI API features are currently supported when using NeMo Guardrails:
+
+| Feature | Status | Notes |
+| :------ | :----: | :---- |
+| **Basic Chat Completion** | ✔ Supported | Full support for standard chat completions with guardrails applied |
+| **Streaming Responses** | ✔ Supported | Server-Sent Events (SSE) streaming with `stream=true` |
+| **Multimodal Input** | ✖ Unsupported | Support for text and image inputs (vision models) with guardrails but not yet OpenAI compatible  |
+| **Function Calling** | ✖ Unsupported | Not yet implemented; guardrails need structured output support |
+| **Tools** | ✖ Unsupported | Related to function calling; requires action flow integration |
+| **Response Format (JSON Mode)** | ✖ Unsupported | Structured output with guardrails requires additional validation logic |

nemoguardrails/colang/v2_x/runtime/runtime.py

@@ -31,6 +31,7 @@
     ColangSyntaxError,
 )
 from nemoguardrails.colang.v2_x.runtime.flows import Event, FlowStatus
+from nemoguardrails.colang.v2_x.runtime.serialization import json_to_state
 from nemoguardrails.colang.v2_x.runtime.statemachine import (
     FlowConfig,
     InternalEvent,
@@ -394,10 +395,13 @@ async def process_events(
             state = State(flow_states={}, flow_configs=self.flow_configs, rails_config=self.config)
             initialize_state(state)
         elif isinstance(state, dict):
-            # TODO: Implement dict to State conversion
-            raise NotImplementedError()
-        #     if isinstance(state, dict):
-        #         state = State.from_dict(state)
+            # Convert dict to State object
+            if state.get("version") == "2.x" and "state" in state:
+                # Handle the serialized state format from API calls
+                state = json_to_state(state["state"])
+            else:
+                # TODO: Implement other dict to State conversion formats if needed
+                raise NotImplementedError("Unsupported state dict format")
 
         assert isinstance(state, State)
         assert state.main_flow_state is not None

nemoguardrails/rails/llm/llmrails.py

@@ -490,6 +490,11 @@ def _init_llms(self):
                     if not self.llm:
                         self.llm = llm_model
                         self.runtime.register_action_param("llm", self.llm)
+                        self._configure_main_llm_streaming(
+                            self.llm,
+                            model_name=llm_config.model,
+                            provider_name=llm_config.engine,
+                        )
                 else:
                     model_name = f"{llm_config.type}_llm"
                     if not hasattr(self, model_name):