Remove unnecessary images conversion to png in binary operator agent.

It's handled by the ai model interaction handlers in khoj server core.

src/khoj/processor/operator/grounding_agent.py

@@ -9,7 +9,7 @@ from khoj.processor.conversation.utils import construct_structured_message
 from khoj.processor.operator.operator_actions import *
 from khoj.processor.operator.operator_agent_base import AgentActResult
 from khoj.processor.operator.operator_environment_base import EnvState
-from khoj.utils.helpers import convert_image_to_png, get_chat_usage_metrics
+from khoj.utils.helpers import get_chat_usage_metrics
 
 logger = logging.getLogger(__name__)
 
@@ -256,7 +256,7 @@ back() # Use this to go back to the previous page.
 
         # Construct grounding LLM input (using only the latest user prompt + image)
         # We don't pass the full history here, as grounding depends on the *current* state + NL action
-        screenshots = [f"data:image/png;base64,{convert_image_to_png(current_state.screenshot)}"]
+        screenshots = [f"data:image/webp;base64,{current_state.screenshot}"]
         grounding_messages_content = construct_structured_message(
             grounding_user_prompt, screenshots, self.model.name, vision_enabled=True
         )

src/khoj/processor/operator/grounding_agent_uitars.py

@@ -19,7 +19,7 @@ from PIL import Image
 
 from khoj.processor.operator.operator_actions import *
 from khoj.processor.operator.operator_environment_base import EnvState
-from khoj.utils.helpers import convert_image_to_png, get_chat_usage_metrics
+from khoj.utils.helpers import get_chat_usage_metrics
 
 logger = logging.getLogger(__name__)
 
@@ -257,11 +257,8 @@ class GroundingAgentUitars:
             self.thoughts
         ), "The number of observations and actions should be the same."
 
-        screenshot = convert_image_to_png(env_state.screenshot)
+        self.history_images.append(base64.b64decode(env_state.screenshot))
-        self.history_images.append(base64.b64decode(screenshot))
+        self.observations.append({"screenshot": env_state.screenshot, "accessibility_tree": None})
-
-        base64_image = screenshot
-        self.observations.append({"screenshot": base64_image, "accessibility_tree": None})
 
         user_prompt = self.prompt_template.format(
             instruction=instruction, action_space=self.prompt_action_space, language=self.language

src/khoj/processor/operator/operator_agent_binary.py

@@ -17,11 +17,7 @@ from khoj.processor.operator.operator_agent_base import (
 )
 from khoj.processor.operator.operator_environment_base import EnvState, EnvStepResult
 from khoj.routers.helpers import send_message_to_model_wrapper
-from khoj.utils.helpers import (
+from khoj.utils.helpers import get_openai_async_client, is_none_or_empty
-    convert_image_to_png,
-    get_openai_async_client,
-    is_none_or_empty,
-)
 
 logger = logging.getLogger(__name__)
 
@@ -132,7 +128,7 @@ Focus on the visual action and provide all necessary context.
 
         if is_none_or_empty(self.messages):
             query_text = f"**Main Objective**: {self.query}"
-            query_screenshot = [f"data:image/png;base64,{convert_image_to_png(current_state.screenshot)}"]
+            query_screenshot = [f"data:image/webp;base64,{current_state.screenshot}"]
             first_message_content = construct_structured_message(
                 message=query_text,
                 images=query_screenshot,

src/khoj/utils/helpers.py

@@ -598,20 +598,6 @@ def convert_image_data_uri(image_data_uri: str, target_format: str = "png") -> s
         return output_data_uri
 
 
-def convert_image_to_png(image_base64: str) -> str:
-    """Convert base64 image to png format for wider support"""
-    image_bytes = base64.b64decode(image_base64)
-    image_io = io.BytesIO(image_bytes)
-    with Image.open(image_io) as original_image:
-        output_image_io = io.BytesIO()
-        original_image.save(output_image_io, "PNG")
-
-        # Encode the WebP image back to base64
-        output_image_bytes = output_image_io.getvalue()
-        output_image_io.close()
-        return base64.b64encode(output_image_bytes).decode("utf-8")
-
-
 def truncate_code_context(original_code_results: dict[str, Any], max_chars=10000) -> dict[str, Any]:
     """
     Truncate large output files and drop image file data from code results.