orgo 0.0.11__tar.gz → 0.0.12__tar.gz

{orgo-0.0.11 → orgo-0.0.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: orgo
-Version: 0.0.11
+Version: 0.0.12
 Summary: Computers for AI agents
 Author: Orgo Team
 License: MIT

{orgo-0.0.11 → orgo-0.0.12}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "orgo"
-version = "0.0.11"
+version = "0.0.12"
 description = "Computers for AI agents"
 authors = [{name = "Orgo Team"}]
 license = {text = "MIT"}

{orgo-0.0.11 → orgo-0.0.12}/src/orgo/prompt.py

@@ -3,6 +3,7 @@ Prompt module for interacting with virtual computers using AI models.
 """
 
 import os
+import base64
 from typing import Dict, List, Any, Optional, Callable, Union, Protocol
 
 
@@ -105,7 +106,14 @@ class AnthropicProvider:
   - File browser items: DOUBLE-CLICK to open folders and files
   - When submitting, use the 'Enter' key, not the 'Return' key.
 * If you see an icon on the desktop that you need to open, ALWAYS use the double_click action, never use left_click.
-</UBUNTU_DESKTOP_GUIDELINES>"""
+</UBUNTU_DESKTOP_GUIDELINES>
+
+<SCREENSHOT_GUIDELINES>
+* Be mindful of how many screenshots you take - they consume significant memory.
+* Only take screenshots when you need to see the current state of the screen.
+* Try to batch multiple actions before taking another screenshot.
+* For better performance, limit the number of screenshots you take.
+</SCREENSHOT_GUIDELINES>"""
         
         try:
             # Define the computer tool per Anthropic's documentation
@@ -138,9 +146,9 @@ class AnthropicProvider:
             while iteration < max_iterations:
                 iteration += 1
                 
-                # Prune old screenshots if we've exceeded our limit
+                # Filter to keep only the N most recent screenshots
                 if screenshot_count > max_saved_screenshots:
-                    self._prune_old_screenshots(messages, screenshot_count - max_saved_screenshots)
+                    self._filter_to_n_most_recent_images(messages, max_saved_screenshots)
                     screenshot_count = max_saved_screenshots
                 
                 # Create the request parameters
@@ -161,7 +169,20 @@ class AnthropicProvider:
                     }
                 
                 # Create message request to Claude
-                response = client.beta.messages.create(**request_params)
+                try:
+                    response = client.beta.messages.create(**request_params)
+                except Exception as e:
+                    if "base64" in str(e).lower():
+                        # If we get a base64 error, try again after more aggressively filtering images
+                        if callback:
+                            callback("error", f"Base64 error detected. Attempting recovery...")
+                        
+                        # Remove all but the most recent image and try again
+                        self._filter_to_n_most_recent_images(messages, 1)
+                        response = client.beta.messages.create(**request_params)
+                    else:
+                        # Not a base64 error, re-raise
+                        raise
                 
                 # Extract the content from the response
                 response_content = response.content
@@ -226,49 +247,57 @@ class AnthropicProvider:
             if callback:
                 callback("error", str(e))
             raise
-            
-    def _prune_old_screenshots(self, messages: List[Dict[str, Any]], num_to_prune: int):
+    
+    def _filter_to_n_most_recent_images(self, messages: List[Dict[str, Any]], max_images: int):
         """
-        Remove old screenshots from the conversation history.
+        Keep only the N most recent images in the conversation history.
         
         Args:
             messages: The conversation history
-            num_to_prune: Number of screenshots to remove
+            max_images: Maximum number of images to keep
         """
-        screenshots_pruned = 0
+        # Find all the image blocks in the conversation history
+        image_blocks = []
         
-        # Start from the beginning of the messages (excluding the first user message)
-        for i in range(1, len(messages)):
-            if messages[i]["role"] != "user":
+        for msg_idx, msg in enumerate(messages):
+            if msg["role"] != "user":
                 continue
                 
-            content = messages[i]["content"]
+            content = msg.get("content", [])
             if not isinstance(content, list):
                 continue
                 
-            # Look for tool_result blocks in the content
-            for j, block in enumerate(content):
+            for content_idx, block in enumerate(content):
                 if not isinstance(block, dict):
                     continue
                     
                 if block.get("type") != "tool_result":
                     continue
-                    
-                # Check if this tool_result contains an image
+                
                 block_content = block.get("content", [])
-                for k, content_item in enumerate(block_content):
+                for content_item_idx, content_item in enumerate(block_content):
                     if not isinstance(content_item, dict):
                         continue
                         
-                    if content_item.get("type") == "image":
-                        # This is a screenshot, remove it
-                        if "source" in content_item and "data" in content_item["source"]:
-                            # Replace the base64 data with a placeholder to save space
-                            content_item["source"]["data"] = "[IMAGE DATA REMOVED]"
-                            screenshots_pruned += 1
-                            
-                            if screenshots_pruned >= num_to_prune:
-                                return
+                    if content_item.get("type") == "image" and "source" in content_item:
+                        image_blocks.append({
+                            "msg_idx": msg_idx,
+                            "content_idx": content_idx,
+                            "block": block,
+                            "content_item_idx": content_item_idx,
+                            "content_item": content_item
+                        })
+        
+        # If we have more images than our limit, remove the oldest ones
+        if len(image_blocks) > max_images:
+            # Keep only the most recent ones (which are at the end of the list)
+            images_to_remove = image_blocks[:-max_images]
+            
+            for img_block in images_to_remove:
+                content_item = img_block["content_item"]
+                if "source" in content_item and "data" in content_item["source"]:
+                    # Replace the base64 data with a placeholder
+                    content_item["source"]["data"] = "[IMAGE DATA REMOVED]"
     
     def _execute_tool(self, 
                       computer_id: str,

{orgo-0.0.11 → orgo-0.0.12}/src/orgo.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: orgo
-Version: 0.0.11
+Version: 0.0.12
 Summary: Computers for AI agents
 Author: Orgo Team
 License: MIT