feat: support OpenAI Flex Processing via [litellm] extra_body config

docs/docs/usage-guide/changing_a_model.md

@@ -32,6 +32,16 @@ OPENAI__API_BASE=https://api.openai.com/v1
 OPENAI__KEY=sk-...
 ```
 
+### OpenAI Flex Processing
+
+To reduce costs for non-urgent/background tasks, enable Flex Processing:
+
+```toml
+[litellm]
+extra_body='{"processing_mode": "flex"}'
+```
+
+See [OpenAI Flex Processing docs](https://platform.openai.com/docs/guides/flex-processing) for details.
 
 ### Azure
 

pr_agent/algo/ai_handlers/litellm_ai_handler.py

@@ -175,6 +175,37 @@ def prepare_logs(self, response, system, user, resp, finish_reason):
             response_log['main_pr_language'] = 'unknown'
         return response_log
 
+    def _process_litellm_extra_body(self, kwargs: dict) -> dict:
+        """
+        Process LITELLM.EXTRA_BODY configuration and update kwargs accordingly.
+
+        Args:
+            kwargs: The current kwargs dictionary to update
+
+        Returns:
+            Updated kwargs dictionary
+
+        Raises:
+            ValueError: If extra_body contains invalid JSON, unsupported keys, or colliding keys
+        """
+        allowed_extra_body_keys = {"processing_mode", "service_tier"}
+        extra_body = getattr(getattr(get_settings(), "litellm", None), "extra_body", None)
+        if extra_body:
+            try:
+                litellm_extra_body = json.loads(extra_body)
+                if not isinstance(litellm_extra_body, dict):
+                    raise ValueError("LITELLM.EXTRA_BODY must be a JSON object")
+                unsupported_keys = set(litellm_extra_body.keys()) - allowed_extra_body_keys
+                if unsupported_keys:
+                    raise ValueError(f"LITELLM.EXTRA_BODY contains unsupported keys: {', '.join(unsupported_keys)}. Allowed keys: {', '.join(allowed_extra_body_keys)}")
+                colliding_keys = kwargs.keys() & litellm_extra_body.keys()
+                if colliding_keys:
+                    raise ValueError(f"LITELLM.EXTRA_BODY cannot override existing parameters: {', '.join(colliding_keys)}")
+                kwargs.update(litellm_extra_body)
+            except json.JSONDecodeError as e:
+                raise ValueError(f"LITELLM.EXTRA_BODY contains invalid JSON: {str(e)}")
+        return kwargs
+
     def _configure_claude_extended_thinking(self, model: str, kwargs: dict) -> dict:
         """
         Configure Claude extended thinking parameters if applicable.
@@ -364,6 +395,9 @@ async def chat_completion(self, model: str, system: str, user: str, temperature:
                     raise ValueError(f"LITELLM.EXTRA_HEADERS contains invalid JSON: {str(e)}")
                 kwargs["extra_headers"] = litellm_extra_headers
 
+            # Support for custom OpenAI body fields (e.g., Flex Processing)
+            kwargs = self._process_litellm_extra_body(kwargs)
+
             get_logger().debug("Prompts", artifact={"system": system, "user": user})
 
             if get_settings().config.verbosity_level >= 2:

pr_agent/settings/.secrets_template.toml

@@ -16,6 +16,10 @@ key = ""  # Acquire through https://platform.openai.com
 #deployment_id = ""  # The deployment name you chose when you deployed the engine
 #fallback_deployments = []  # For each fallback model specified in configuration.toml in the [config] section, specify the appropriate deployment_id
 
+# OpenAI Flex Processing (optional, for cost savings)
+# [litellm]
+# extra_body='{"processing_mode": "flex"}'
+
 [pinecone]
 api_key = "..."
 environment = "gcp-starter"