✨ Add logprobs param to llm if its config has been added

backend/agents/create_agent_info.py

@@ -74,6 +74,7 @@
     AIDP_TENANT_ID,
     DATA_PROCESS_SERVICE,
     LANGUAGE,
+    LLM_INCLUDE_LOGPROBS,
     LOCAL_MCP_SERVER,
     MINIO_DEFAULT_BUCKET,
     MODEL_CONFIG_MAPPING,
@@ -856,6 +857,7 @@ def _get_skill_script_tools(
 async def create_model_config_list(tenant_id):
     records = get_model_records({"model_type": "llm"}, tenant_id)
     model_list = []
+    extra_body = {"logprobs": True} if LLM_INCLUDE_LOGPROBS else None
     for record in records:
         model_list.append(
             ModelConfig(cite_name=record["display_name"],
@@ -880,7 +882,8 @@ async def create_model_config_list(tenant_id):
                         default_output_reserve_tokens=record.get("default_output_reserve_tokens"),
                         tokenizer_family=record.get("tokenizer_family"),
                         capacity_source=record.get("capacity_source"),
-                        capability_profile_version=record.get("capability_profile_version")))
+                        capability_profile_version=record.get("capability_profile_version"),
+                        extra_body=extra_body))
     # fit for old version, main_model and sub_model use default model
     main_model_config = tenant_config_manager.get_model_config(
         key=MODEL_CONFIG_MAPPING["llm"], tenant_id=tenant_id)
@@ -896,7 +899,8 @@ async def create_model_config_list(tenant_id):
                     model_factory=main_model_config.get("model_factory"),
                     timeout_seconds=main_model_config.get("timeout_seconds"),
                     concurrency_limit=main_model_config.get("concurrency_limit"),
-                    prompt_cache=main_prompt_cache))
+                    prompt_cache=main_prompt_cache,
+                    extra_body=extra_body))
     model_list.append(
         ModelConfig(cite_name="sub_model",
                     api_key=main_model_config.get("api_key", ""),
@@ -907,7 +911,8 @@ async def create_model_config_list(tenant_id):
                     model_factory=main_model_config.get("model_factory"),
                     timeout_seconds=main_model_config.get("timeout_seconds"),
                     concurrency_limit=main_model_config.get("concurrency_limit"),
-                    prompt_cache=main_prompt_cache))
+                    prompt_cache=main_prompt_cache,
+                    extra_body=extra_body))
 
     return model_list
 

backend/consts/const.py

@@ -756,5 +756,10 @@ def _resolve_app_version(default: str = "v2.2.1") -> str:
     "execution_logs",
 ])
 
+# LLM Model Configuration
+LLM_INCLUDE_LOGPROBS = os.getenv("LLM_INCLUDE_LOGPROBS", "false").lower() == "true"
+"""When True, adds logprobs=true to every chat.completions.create request body,
+enabling the provider to return log probability information in the response."""
+
 # SSE streaming event type for status messages
 STREAM_STATUS_EVENT = "event: stream_status\n"

deploy/env/.env.example

@@ -150,6 +150,11 @@ DISABLE_CELERY_FLOWER=true
 DOCKER_ENVIRONMENT=false
 ENABLE_UPLOAD_IMAGE=false
 
+# LLM Model Configuration
+# When true, adds logprobs=true to every chat.completions.create request body,
+# enabling the provider to return log probability information in the response.
+# Default: false (logprobs disabled, backward-compatible behaviour).
+LLM_INCLUDE_LOGPROBS=false
 
 # Celery Configuration
 CELERY_WORKER_PREFETCH_MULTIPLIER=1

sdk/nexent/core/models/openai_llm.py

@@ -623,6 +623,15 @@ def _dispatch_chat_completion(
                     caller_value=caller_max_tokens,
                 )
             completion_kwargs["max_tokens"] = trusted_max_tokens
+        logger.info(
+            "event=chat_completion_create model_id=%s kwargs=%s",
+            self.model_id,
+            json.dumps(
+                {k: v for k, v in completion_kwargs.items() if k != "messages"},
+                ensure_ascii=False,
+                default=str,
+            ),
+        )
         return self.client.chat.completions.create(**completion_kwargs)
 
     @staticmethod