From 56ed79cf462819a1a4a90e96bc523527b43f3905 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=8F=A4=E4=BB=8A?= Date: Sun, 20 Apr 2025 12:36:25 +0800 Subject: [PATCH] delete unused message in browser when not support image --- app/agent/browser.py | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-) diff --git a/app/agent/browser.py b/app/agent/browser.py index 92d8ea62f..8304d4b6f 100644 --- a/app/agent/browser.py +++ b/app/agent/browser.py @@ -4,6 +4,7 @@ from pydantic import Field, model_validator from app.agent.toolcall import ToolCallAgent +from app.llm import MULTIMODAL_MODELS from app.logger import logger from app.prompt.browser import NEXT_STEP_PROMPT, SYSTEM_PROMPT from app.schema import Message, ToolChoice @@ -57,12 +58,15 @@ async def format_next_step_prompt(self) -> str: if pixels_below > 0: content_below_info = f" ({pixels_below} pixels)" + supports_images = self.agent.llm.model in MULTIMODAL_MODELS + if self._current_base64_image: - image_message = Message.user_message( - content="Current browser screenshot:", - base64_image=self._current_base64_image, - ) - self.agent.memory.add_message(image_message) + if supports_images: + image_message = Message.user_message( + content="Current browser screenshot:", + base64_image=self._current_base64_image, + ) + self.agent.memory.add_message(image_message) self._current_base64_image = None # Consume the image after adding return NEXT_STEP_PROMPT.format(