url-09: fetched images become vision input — fetch_url og:image/body images ride as input_image, direct image urls ingested, data urls never in json tool text

fjerkroa_bot/openai_responder.py

@@ -329,6 +329,17 @@ class OpenAIResponder(AIResponder, LeonardoAIDrawMixIn):
             items.append(data)
         return items
 
+    @staticmethod
+    def _split_vision(function_result: Any) -> Tuple[Any, List[str]]:
+        """Detach cached image data URLs from a tool result (URL-09) — they
+        ride to the model as image input, never as JSON text (a base64 data
+        URL would blow the 8000-char sanitizer cap)."""
+        if isinstance(function_result, dict) and function_result.get("vision"):
+            return function_result, [str(url) for url in function_result.pop("vision")]
+        if isinstance(function_result, dict):
+            function_result.pop("vision", None)
+        return function_result, []
+
     @staticmethod
     def _responses_refused(result: Any) -> bool:
         for item in getattr(result, "output", []) or []:
@@ -381,6 +392,7 @@ class OpenAIResponder(AIResponder, LeonardoAIDrawMixIn):
                 function_args = json.loads(call.arguments) if call.arguments else {}
                 logging.info(f"🔧 Executing tool: {call.name} with args: {function_args}")
                 function_result = await self._dispatch_tool(call.name, function_args, author or "")
+                function_result, vision = self._split_vision(function_result)
                 logging.info(f"🔧 Tool result: {type(function_result)} - {str(function_result)[:200]}...")
                 context.append(
                     {
@@ -390,6 +402,10 @@ class OpenAIResponder(AIResponder, LeonardoAIDrawMixIn):
                         "output": sanitize_external_text(json.dumps(function_result), 8000) if function_result else "No results found",
                     }
                 )
+                if vision:
+                    # fetched images become sight, not text (URL-09)
+                    logging.info(f"🔧 Tool returned {len(vision)} image(s) — attached as vision input")
+                    context.append({"role": "user", "content": [{"type": "input_image", "image_url": url} for url in vision]})
             kwargs["input"] = context
             rounds -= 1
             if rounds <= 0:
@@ -504,6 +520,7 @@ class OpenAIResponder(AIResponder, LeonardoAIDrawMixIn):
 
                         # Route to the right provider (IGDB or URL reader)
                         function_result = await self._dispatch_tool(function_name, function_args, self._last_author(messages) or "")
+                        function_result, vision = self._split_vision(function_result)
 
                         logging.info(f"🔧 Tool result: {type(function_result)} - {str(function_result)[:200]}...")
 
@@ -517,6 +534,12 @@ class OpenAIResponder(AIResponder, LeonardoAIDrawMixIn):
                                 ),
                             }
                         )
+                        if vision:
+                            # fetched images become sight, not text (URL-09)
+                            logging.info(f"🔧 Tool returned {len(vision)} image(s) — attached as vision input")
+                            messages.append(
+                                {"role": "user", "content": [{"type": "image_url", "image_url": {"url": url}} for url in vision]}
+                            )
 
                     # Get final response after function execution - remove tools for final call
                     final_chat_kwargs = {