From 4fd9d6b6d5b741ede72439ddec6498eb021e8bd3 Mon Sep 17 00:00:00 2001
From: busya <uglypod@gmail.com>
Date: Tue, 24 Feb 2026 14:22:08 +0300
Subject: [PATCH] fix(llm): skip unsupported json_object mode for openrouter
 stepfun models

---
 backend/src/plugins/llm_analysis/service.py | 51 +++++++++++++++++----
 1 file changed, 43 insertions(+), 8 deletions(-)

diff --git a/backend/src/plugins/llm_analysis/service.py b/backend/src/plugins/llm_analysis/service.py
index b4b78bc..4ef46af 100644
--- a/backend/src/plugins/llm_analysis/service.py
+++ b/backend/src/plugins/llm_analysis/service.py
@@ -437,6 +437,26 @@ class LLMClient:
         self.client = AsyncOpenAI(api_key=api_key, base_url=base_url)
     # [/DEF:LLMClient.__init__:Function]
 
+    # [DEF:LLMClient._supports_json_response_format:Function]
+    # @PURPOSE: Detect whether provider/model is likely compatible with response_format=json_object.
+    # @PRE: Client initialized with base_url and default_model.
+    # @POST: Returns False for known-incompatible combinations to avoid avoidable 400 errors.
+    def _supports_json_response_format(self) -> bool:
+        base = (self.base_url or "").lower()
+        model = (self.default_model or "").lower()
+
+        # OpenRouter routes to many upstream providers; some models reject json_object mode.
+        if "openrouter.ai" in base:
+            incompatible_tokens = (
+                "stepfun/",
+                "step-",
+                ":free",
+            )
+            if any(token in model for token in incompatible_tokens):
+                return False
+        return True
+    # [/DEF:LLMClient._supports_json_response_format:Function]
+
     # [DEF:LLMClient.get_json_completion:Function]
     # @PURPOSE: Helper to handle LLM calls with JSON mode and fallback parsing.
     # @PRE: messages is a list of valid message dictionaries.
@@ -460,19 +480,34 @@ class LLMClient:
         with belief_scope("get_json_completion"):
             response = None
             try:
+                use_json_mode = self._supports_json_response_format()
                 try:
-                    logger.info(f"[get_json_completion] Attempting LLM call with JSON mode for model: {self.default_model}")
+                    logger.info(
+                        f"[get_json_completion] Attempting LLM call for model: {self.default_model} "
+                        f"(json_mode={'on' if use_json_mode else 'off'})"
+                    )
                     logger.info(f"[get_json_completion] Base URL being used: {self.base_url}")
                     logger.info(f"[get_json_completion] Number of messages: {len(messages)}")
                     logger.info(f"[get_json_completion] API Key present: {bool(self.api_key and len(self.api_key) > 0)}")
-                    
-                    response = await self.client.chat.completions.create(
-                        model=self.default_model,
-                        messages=messages,
-                        response_format={"type": "json_object"}
-                    )
+
+                    if use_json_mode:
+                        response = await self.client.chat.completions.create(
+                            model=self.default_model,
+                            messages=messages,
+                            response_format={"type": "json_object"}
+                        )
+                    else:
+                        response = await self.client.chat.completions.create(
+                            model=self.default_model,
+                            messages=messages
+                        )
                 except Exception as e:
-                    if "JSON mode is not enabled" in str(e) or "400" in str(e):
+                    if use_json_mode and (
+                        "JSON mode is not enabled" in str(e)
+                        or "json_object is not supported" in str(e).lower()
+                        or "response_format" in str(e).lower()
+                        or "400" in str(e)
+                    ):
                         logger.warning(f"[get_json_completion] JSON mode failed or not supported: {str(e)}. Falling back to plain text response.")
                         response = await self.client.chat.completions.create(
                             model=self.default_model,