versionhq 1.2.4.7__tar.gz → 1.2.4.9__tar.gz

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/.github/workflows/run_tests.yml

@@ -45,6 +45,7 @@ jobs:
           uv pip install --upgrade pip pytest
           uv lock --upgrade
           uv sync --all-extras --no-extra pygraphviz
+          playwright install
 
       - name: Pytest
         run: uv run pytest tests -vv --cache-clear

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.2
+Metadata-Version: 2.4
 Name: versionhq
-Version: 1.2.4.7
+Version: 1.2.4.9
 Summary: Autonomous agent networks for task automation with multi-step reasoning.
 Author-email: Kuriko Iwai <kuriko@versi0n.io>
 License: MIT License
@@ -45,7 +45,6 @@ Requires-Python: >=3.11
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: regex==2024.11.6
-Requires-Dist: requests>=2.32.3
 Requires-Dist: pydantic>=2.10.6
 Requires-Dist: typing>=0.0.0
 Requires-Dist: json-repair>=0.0.0
@@ -77,9 +76,11 @@ Requires-Dist: pygraphviz>=1.14; extra == "pygraphviz"
 Provides-Extra: tools
 Requires-Dist: html2text>=2024.2.26; extra == "tools"
 Requires-Dist: sec-api>=1.0.28; extra == "tools"
+Requires-Dist: pytest-playwright>=0.7.0; extra == "tools"
 Provides-Extra: torch
 Requires-Dist: torch>=2.6.0; extra == "torch"
 Requires-Dist: torchvision>=0.21.0; extra == "torch"
+Dynamic: license-file
 
 # Overview
 
@@ -208,7 +209,7 @@ A `TaskGraph` represents tasks as `nodes` and their execution dependencies as `e
 
 `Agent Networks` can handle `TaskGraph` objects by optimizing their formations.
 
-<img src="https://res.cloudinary.com/dfeirxlea/image/upload/v1739337639/pj_m_home/zfg4ccw1m1ww1tpnb0pa.png">
+<img src="https://res.cloudinary.com/dfeirxlea/image/upload/v1739337639/pj_m_home/zfg4ccw1m1ww1tpnb0pa.png" width="300px">
 
 <hr />
 
@@ -290,8 +291,11 @@ class CustomOutput(BaseModel):
    test1: str
    test2: list[str]
 
-def dummy_func(message: str, test1: str, test2: list[str]) -> str:
-   return f"""{message}: {test1}, {", ".join(test2)}"""
+def dummy_func(message: str, **kwargs) -> str:
+   test1 = kwargs["test1"] if kwargs and "test1" in kwargs else ""
+   test2 = kwargs["test2"] if kwargs and "test2" in kwargs else ""
+   if test1 and test2:
+      return f"""{message}: {test1}, {", ".join(test2)}"""
 
 task = vhq.Task(
    description="Amazing task",

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/README.md

@@ -125,7 +125,7 @@ A `TaskGraph` represents tasks as `nodes` and their execution dependencies as `e
 
 `Agent Networks` can handle `TaskGraph` objects by optimizing their formations.
 
-<img src="https://res.cloudinary.com/dfeirxlea/image/upload/v1739337639/pj_m_home/zfg4ccw1m1ww1tpnb0pa.png">
+<img src="https://res.cloudinary.com/dfeirxlea/image/upload/v1739337639/pj_m_home/zfg4ccw1m1ww1tpnb0pa.png" width="300px">
 
 <hr />
 
@@ -207,8 +207,11 @@ class CustomOutput(BaseModel):
    test1: str
    test2: list[str]
 
-def dummy_func(message: str, test1: str, test2: list[str]) -> str:
-   return f"""{message}: {test1}, {", ".join(test2)}"""
+def dummy_func(message: str, **kwargs) -> str:
+   test1 = kwargs["test1"] if kwargs and "test1" in kwargs else ""
+   test2 = kwargs["test2"] if kwargs and "test2" in kwargs else ""
+   if test1 and test2:
+      return f"""{message}: {test1}, {", ".join(test2)}"""
 
 task = vhq.Task(
    description="Amazing task",

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/docs/core/task/task-execution.md

@@ -192,7 +192,7 @@ Callback results will be stored in `callback_output` filed of the `TaskOutput` o
 ```python
 import versionhq as vhq
 
-def callback_func(condition: str, test1: str):
+def callback_func(condition: str, test1: str, **kwargs):
     return f"Result: {test1}, condition added: {condition}"
 
 task = vhq.Task(

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/docs/core/task/task-output.md

@@ -22,8 +22,13 @@ class CustomOutput(BaseModel):
 def dummy_tool():
     return "dummy"
 
-def summarize_response(message: str, test1: str, test2: list[str]) -> str:
-    return f"""{message}: {test1}, {", ".join(test2)}"""
+def summarize_response(message: str, **kwargs) -> str | None:
+    test1 = kwargs["test1"] if kwargs and "test1" in kwargs else None
+    test2 = kwargs["test2"] if kwargs and "test2" in kwargs else None
+    if test1 and test2:
+        return f"""{message}: {test1}, {", ".join(str(test2))}"""
+    else:
+        return None
 
 task = vhq.Task(
     description="Research a topic to teach a kid aged 6 about math.",
@@ -41,8 +46,8 @@ assert res.task_id == task.id
 assert res.raw
 assert res.json_dict
 assert res.pydantic.test1 and res.pydantic.test2
-assert "Hi! Here is the result: " in res.callback_output
-assert res.pydantic.test1 in res.callback_output and ", ".join(res.pydantic.test2) in res.callback_output
+if res.callback_output:
+    assert "Hi! Here is the result: " in res.callback_output
 assert res.tool_output is None
 assert res.evaluation and isinstance(res.evaluation, vhq.Evaluation)
 ```

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/docs/core/task/task-strc-response.md

@@ -177,10 +177,13 @@ class Main(BaseModel):
     main1: list[Any] # <= assume expecting to store Sub object.
     main2: dict[str, Any]
 
-def format_response(sub, main1, main2) -> Main:
+
+def format_response(sub, **kwargs) -> Main:
+    main1 = kwargs["main1"] if kwargs and "main1" in kwargs else None
     if main1:
         main1.append(sub)
-    main = Main(main1=main1, main2=main2)
+    main2 = kwargs["main2"] if kwargs and "main2" in kwargs else None
+    main = Main(main1=main1, main2=str(main2))
     return main
 
 # 3. Executes

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/docs/quickstart.md

@@ -44,9 +44,11 @@ class CustomOutput(BaseModel):
    test1: str
    test2: list[str]
 
-def dummy_func(message: str, test1: str, test2: list[str]) -> str:
-   return f"""{message}: {test1}, {", ".join(test2)}"""
-
+def dummy_func(message: str, **kwargs) -> str:
+   test1 = kwargs["test1"] if kwargs and "test1" in kwargs else ""
+   test2 = kwargs["test2"] if kwargs and "test2" in kwargs else ""
+   if test1 and test2:
+      return f"""{message}: {test1}, {", ".join(test2)}"""
 
 agent = vhq.Agent(role="demo manager")
 

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/pyproject.toml

@@ -15,7 +15,7 @@ exclude = ["test*", "__pycache__", "*.egg-info"]
 
 [project]
 name = "versionhq"
-version = "1.2.4.7"
+version = "1.2.4.9"
 authors = [{ name = "Kuriko Iwai", email = "kuriko@versi0n.io" }]
 description = "Autonomous agent networks for task automation with multi-step reasoning."
 readme = "README.md"
@@ -24,7 +24,6 @@ license = { file = "LICENSE" }
 keywords = ["autonomic agent networks", "deep agent", "agentic orchestration framework", ]
 dependencies = [
   "regex==2024.11.6",
-  "requests>=2.32.3",
   "pydantic>=2.10.6",
   "typing>=0.0.0",
   "json-repair>=0.0.0",
@@ -82,6 +81,7 @@ pygraphviz = [
 tools = [
     "html2text>=2024.2.26",
     "sec-api>=1.0.28",
+    "pytest-playwright>=0.7.0",
 ]
 torch = [
     "torch>=2.6.0",

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/requirements.txt

@@ -1,5 +1,4 @@
 regex==2024.11.6
-requests>=2.32.3
 pydantic>=2.10.6
 typing
 json-repair

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/src/versionhq/__init__.py

@@ -25,7 +25,7 @@ from versionhq.tool.rag_tool import RagTool
 from versionhq.tool.cache_handler import CacheHandler
 from versionhq.tool.tool_handler import ToolHandler
 from versionhq.tool.composio.model import ComposioBaseTool
-from versionhq.tool.gpt.cup import GPTToolCUP, CUPToolSchema
+from versionhq.tool.gpt.cua import GPTToolCUA, CUAToolSchema
 from versionhq.tool.gpt.file_search import GPTToolFileSearch, FilterSchema
 from versionhq.tool.gpt.web_search import GPTToolWebSearch
 from versionhq.memory.contextual_memory import ContextualMemory
@@ -35,7 +35,7 @@ from versionhq.agent_network.formation import form_agent_network
 from versionhq.task_graph.draft import workflow
 
 
-__version__ = "1.2.4.7"
+__version__ = "1.2.4.9"
 __all__ = [
     "Agent",
 
@@ -90,8 +90,8 @@ __all__ = [
     "ToolHandler",
     "ComposioBaseTool",
 
-    "GPTToolCUP",
-    "CUPToolSchema",
+    "GPTToolCUA",
+    "CUAToolSchema",
     "GPTToolFileSearch",
     "FilterSchema",
     "GPTToolWebSearch",

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/src/versionhq/_utils/logger.py

@@ -51,7 +51,7 @@ class Logger(BaseModel):
     def log(self, level: str, message: str, color="yellow"):
         if self.verbose:
             timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-            self._printer.print(f"\n{timestamp} - versionHQ [{level.upper()}]: {message}", color=color)
+            self._printer.print(f"\n{timestamp} - vhq [{level.upper()}]: {message}", color=color)
 
         self._save(level=level, message=message, filename=self.filename)
 

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/src/versionhq/agent/model.py

@@ -126,7 +126,7 @@ class Agent(BaseModel):
         from versionhq.tool.rag_tool import RagTool
         from versionhq.tool.gpt.web_search import GPTToolWebSearch
         from versionhq.tool.gpt.file_search import GPTToolFileSearch
-        from versionhq.tool.gpt.cup import GPTToolCUP
+        from versionhq.tool.gpt.cua import GPTToolCUA
 
         if not self.tools:
             return self
@@ -134,7 +134,7 @@ class Agent(BaseModel):
         tool_list = []
         for item in self.tools:
             match item:
-                case RagTool() | BaseTool() | GPTToolCUP() | GPTToolFileSearch() | GPTToolWebSearch():
+                case RagTool() | BaseTool() | GPTToolCUA() | GPTToolFileSearch() | GPTToolWebSearch():
                     tool_list.append(item)
 
                 case Tool():
@@ -433,7 +433,7 @@ class Agent(BaseModel):
         from versionhq.tool.rag_tool import RagTool
         from versionhq.tool.gpt.web_search import GPTToolWebSearch
         from versionhq.tool.gpt.file_search import GPTToolFileSearch
-        from versionhq.tool.gpt.cup import GPTToolCUP
+        from versionhq.tool.gpt.cua import GPTToolCUA
 
         all_tools = []
         if task: all_tools = task.tools + self.tools if task.can_use_agent_tools else task.tools
@@ -446,7 +446,7 @@ class Agent(BaseModel):
                     case RagTool():
                         rag_tools.append(item)
 
-                    case GPTToolCUP() | GPTToolFileSearch() | GPTToolWebSearch():
+                    case GPTToolCUA() | GPTToolFileSearch() | GPTToolWebSearch():
                         gpt_tools.append(item)
 
                     case Tool() | BaseTool() | ToolSet():

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/src/versionhq/task/model.py

@@ -18,7 +18,7 @@ from versionhq.tool.model import Tool, ToolSet, BaseTool
 from versionhq.tool.rag_tool import RagTool
 from versionhq.tool.gpt.web_search import GPTToolWebSearch
 from versionhq.tool.gpt.file_search import GPTToolFileSearch
-from versionhq.tool.gpt.cup import GPTToolCUP
+from versionhq.tool.gpt.cua import GPTToolCUA
 from versionhq._utils import process_config, Logger, UsageMetrics, ErrorType
 
 
@@ -374,7 +374,7 @@ class Task(BaseModel):
             tool_list = []
             for item in self.tools:
                 match item:
-                    case Tool() | ToolSet() | BaseTool() | RagTool() | GPTToolCUP() | GPTToolFileSearch() | GPTToolWebSearch():
+                    case Tool() | ToolSet() | BaseTool() | RagTool() | GPTToolCUA() | GPTToolFileSearch() | GPTToolWebSearch():
                         tool_list.append(item)
                     case type(item, callable):
                         tool_list.append(Tool(func=item))
@@ -387,6 +387,8 @@ class Task(BaseModel):
                                 tool = RagTool(**item)
                             except:
                                 pass
+                        if tool:
+                            tool_list.append(tool)
                     case _:
                         pass
             self.tools = tool_list
@@ -713,7 +715,7 @@ class Task(BaseModel):
                 sig = inspect.signature(self.callback)
                 valid_keys = [param.name for param in sig.parameters.values() if param.kind == param.POSITIONAL_OR_KEYWORD]
                 valid_kwargs = { k: kwargs[k] if  k in kwargs else None for k in valid_keys }
-                callback_res = self.callback(**valid_kwargs)
+                callback_res = self.callback(**valid_kwargs, **task_output.json_dict )
                 task_output.callback_output = callback_res
 
         end_dt = datetime.datetime.now()

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/src/versionhq/tool/gpt/_enum.py

@@ -7,14 +7,14 @@ class GPTSizeEnum(str, Enum):
     HIGH = "high"
 
 
-class GPTCUPEnvironmentEnum(str, Enum):
+class GPTCUAEnvironmentEnum(str, Enum):
     BROWSER = "browser"
     MAC = "mac"
     WINDOWS = "windows"
     UNBUNTU = "ubuntu"
 
 
-class GPTCUPTypeEnum(str, Enum):
+class GPTCUATypeEnum(str, Enum):
     COMPUTER_CALL_OUTPUT = "computer_call_output"
     COMPUTER_USE_PREVIEW = "computer_use_preview"
 

versionhq-1.2.4.9/src/versionhq/tool/gpt/cua.py

@@ -0,0 +1,295 @@
+import datetime
+import time
+from typing import List, Dict, Any, Tuple
+
+from versionhq._utils import convert_img_url
+from versionhq.tool.gpt import openai_client
+from versionhq.tool.gpt._enum import GPTCUAEnvironmentEnum, GPTCUATypeEnum, GPTSizeEnum
+from versionhq._utils import is_valid_enum, UsageMetrics, ErrorType, Logger, is_valid_url
+
+
+allowed_browsers = ['webkit', 'chromium', 'firefox']
+
+
+class CUAToolSchema:
+    type: str = GPTCUATypeEnum.COMPUTER_USE_PREVIEW.value
+    display_width: int = 1024
+    display_height: int = 768
+    environment: str = GPTCUAEnvironmentEnum.BROWSER.value
+
+    def __init__(
+            self,
+            type: str | GPTCUATypeEnum = None,
+            display_width: int = None,
+            display_height: int = None,
+            environment: str | GPTCUAEnvironmentEnum = None
+        ):
+        self.display_height = display_height if display_height else self.display_height
+        self.display_width = display_width if display_width else self.display_width
+
+        if type and is_valid_enum(enum=GPTCUATypeEnum, val=type):
+            self.type = type.value if isinstance(type, GPTCUATypeEnum) else type
+
+        if environment and is_valid_enum(enum=GPTCUAEnvironmentEnum, val=environment):
+            self.environment = environment.value if isinstance(environment, GPTCUAEnvironmentEnum) else environment
+
+        self.environment = environment if environment else self.environment
+
+
+    @property
+    def schema(self) -> Dict[str, Any]:
+        return {
+            "type": self.type if isinstance(self.type, str) else self.type.value,
+            "display_width": self.display_width,
+            "display_height": self.display_height,
+            "environment": self.environment if isinstance(self.environment, str) else self.environment.value,
+        }
+
+
+class GPTToolCUA:
+    model: str = "computer-use-preview"
+    tools: List[CUAToolSchema] = list()
+    user_prompt: str = None
+    img_url: str = None
+    web_url: str = "https://www.google.com"
+    browser: str = "firefox"
+    reasoning_effort: str = GPTSizeEnum.MEDIUM.value
+    truncation: str = "auto"
+
+    _response_ids: List[str] = list()
+    _call_ids: List[str] = list()
+    _usage: UsageMetrics = UsageMetrics()
+    _logger: Logger = Logger(info_file_save=True, filename="cua-task-{}".format(str(datetime.datetime.now().timestamp())) + ".png")
+
+
+    def __init__(
+        self,
+        user_prompt: str,
+        tools: List[CUAToolSchema] | CUAToolSchema = None,
+        img_url: str = None,
+        web_url: str = "https://www.google.com",
+        browser: str = "chromium",
+        reasoning_effort: GPTSizeEnum | str = None,
+        truncation: str = None,
+        _usage: UsageMetrics = UsageMetrics()
+    ):
+        self.user_prompt = user_prompt
+        self.web_url = web_url if is_valid_url(web_url) else "https://www.google.com"
+        self.browser = browser if browser in allowed_browsers else 'chromium'
+        self.truncation = truncation if truncation else self.truncation
+        self._usage = _usage
+        self._response_ids = list()
+        self._call_ids = list()
+
+        if img_url:
+            img_url = convert_img_url(img_url)
+            self.img_url = img_url
+
+        if reasoning_effort and is_valid_enum(enum=GPTSizeEnum, val=reasoning_effort):
+            self.reasoning_effort = reasoning_effort.value if isinstance(reasoning_effort, GPTSizeEnum) else reasoning_effort
+
+        if tools:
+            match tools:
+                case list():
+                    if self.tools:
+                        self.tools.extend(tools)
+                    else:
+                        self.tools = tools
+                case CUAToolSchema():
+                    if self.tools:
+                        self.tools.append(tools)
+                    else:
+                        self.tools = [tools]
+                case _:
+                    pass
+
+
+    def _take_screenshot(self, page: Any = None, path: str = None) -> Tuple[str | None, str | None]:
+        import base64
+        if not page:
+            return None, None
+
+        path = path if path else "screenshot.png"
+        screenshot_bytes = page.screenshot()
+        screenshot_base64 = base64.b64encode(screenshot_bytes).decode("utf-8")
+        self._logger.log(message=f"Action: screenshot", level="info", color="blue")
+        return screenshot_bytes, screenshot_base64
+
+
+    def _handle_model_action(self, page: Any, action: Any, action_type: str = None) -> bool:
+        """Creates a page object and performs actions."""
+
+        action_type = action_type if action_type else action.type
+        start_dt = datetime.datetime.now()
+
+        try:
+            match action_type:
+                case "click":
+                    x, y = action.x, action.y
+                    button = action.button
+                    self._logger.log(message=f"Action: click at ({x}, {y}) with button '{button}'", level="info", color="blue")
+                    if button != "left" and button != "right":
+                        button = "left"
+                    page.mouse.click(x, y, button=button)
+
+                case "scroll":
+                    x, y = action.x, action.y
+                    scroll_x, scroll_y = action.scroll_x, action.scroll_y
+                    self._logger.log(message=f"Action: scroll at ({x}, {y}) with offsets (scroll_x={scroll_x}, scroll_y={scroll_y})", level="info", color="blue")
+                    page.mouse.move(x, y)
+                    page.evaluate(f"window.scrollBy({scroll_x}, {scroll_y})")
+
+                case "keypress":
+                    keys = action.keys
+                    for k in keys:
+                        self._logger.log(message=f"Action: keypress '{k}'", level="info", color="blue")
+                        if k.lower() == "enter":
+                            page.keyboard.press("Enter")
+                        elif k.lower() == "space":
+                            page.keyboard.press(" ")
+                        else:
+                            page.keyboard.press(k)
+
+                case "type":
+                    text = action.text
+                    self._logger.log(message=f"Action: type text: {text}", level="info", color="blue")
+                    page.keyboard.type(text)
+
+                case "wait":
+                    self._logger.log(message=f"Action: wait", level="info", color="blue")
+                    time.sleep(2)
+
+                case "screenshot":
+                    pass
+
+                case _:
+                    self._logger.log(message=f"Unrecognized action: {action}", level="warning", color="yellow")
+
+        except Exception as e:
+            self._usage.record_errors(type=ErrorType.API)
+            self._logger.log(message=f"Error handling action {action}: {e}", level="error", color="red")
+
+        end_dt = datetime.datetime.now()
+        self._usage.record_latency(start_dt=start_dt, end_dt=end_dt)
+        return bool(self._usage.total_errors)
+
+
+    def run(self, screenshot: str = None) -> Tuple[Dict[str, Any], None, UsageMetrics]:
+        raw_res = dict()
+        usage = self._usage if self._usage else UsageMetrics()
+        start_dt = datetime.datetime.now()
+
+        try:
+            schema = self.schema
+            if screenshot and "output" in schema["input"][0]:
+                output_image_url = schema["input"][0]["output"]["image_url"].replace("SCREENSHOT", str(screenshot))
+                schema["input"][0]["output"]["image_url"] = output_image_url
+
+            res = openai_client.responses.create(**schema)
+            if not res:
+                usage.record_errors(ErrorType.TOOL)
+            else:
+                for item in res.output:
+                    match item.type:
+                        case "reasoning":
+                            raw_res.update(dict(reasoning=item.summary[0].text))
+                            if item.id and item.id.startwith('rs'):
+                                self._response_ids.append(item.id)
+                        case "computer_call":
+                            raw_res.update(dict(action=item.action))
+                            # self._response_ids.append(item.id)
+                            self._call_ids.append(item.call_id)
+                        case _:
+                            pass
+                usage.record_token_usage(**res.usage.__dict__)
+
+        except Exception as e:
+            self._logger.log(message=f"Failed to run: {str(e)}", color="red", level="error")
+            usage.record_errors(ErrorType.TOOL)
+
+        end_dt = datetime.datetime.now()
+        usage.record_latency(start_dt=start_dt, end_dt=end_dt)
+        return raw_res, None, usage
+
+
+    def invoke_playwright(self) -> Tuple[Dict[str, Any], None, UsageMetrics]:
+        """Handles computer use loop. Ref. OpenAI official website."""
+
+        from playwright.sync_api import sync_playwright
+
+        self._logger.log(message="Start the operation.", level="info", color="blue")
+
+        try:
+            with sync_playwright() as p:
+                b = p.firefox if self.browser == "firefox" else p.webkit if self.browser == "webkit" else p.chromium
+                browser = b.launch(headless=True)
+                page = browser.new_page()
+                if not browser or not page:
+                    return None, None, None
+
+                page.goto(self.web_url)
+                res, _, usage = self.run()
+                self._usage = usage
+                actions = [v for k, v in res.items() if k =="action"] if res else []
+                action = actions[0] if actions else None
+                start_dt = datetime.datetime.now()
+
+                if action:
+                    while True:
+                        self._handle_model_action(page=page, action=action)
+                        _, screenshot_base64 = self._take_screenshot(page=page)
+                        res, _, usage = self.run(screenshot=screenshot_base64)
+                        self._usage.agggregate(metrics=usage)
+                        if not res:
+                            usage.record_errors(type=ErrorType.API)
+                            break
+
+                        actions = [v for k, v in res.items() if k =="action"] if res else []
+                        action = actions[0] if actions else None
+                        if not action:
+                            break
+                else:
+                    self._usage.record_errors(type=ErrorType.TOOL)
+
+        except Exception as e:
+            self._logger.log(message=f"Failed to execute. {str(e)}", color="red", level="error")
+
+        end_dt = datetime.datetime.now()
+        self._usage.record_latency(start_dt=start_dt, end_dt=end_dt)
+        # browser.close()
+        return res, _, self._usage
+
+
+    @property
+    def schema(self) -> Dict[str, Any]:
+        """Formats args schema for CUA calling."""
+
+        tool_schema = [item.schema for item in self.tools]
+        schema = dict()
+        inputs = list()
+        previous_response_id = self._response_ids[-1] if self._response_ids and self._response_ids[-1].startswith("rs") else None
+
+        if self._call_ids:
+            inputs = [
+                {
+                    "call_id": self._call_ids[-1],
+                    "type": "computer_call_output",
+                    "output": { "type": "input_image", "image_url": f"data:image/png;base64,SCREENSHOT"}
+                }
+            ]
+            schema = dict(
+                model=self.model,
+                previous_response_id=previous_response_id,
+                tools=tool_schema,
+                input=inputs,
+                truncation=self.truncation
+            )
+
+        else:
+            img_url = convert_img_url(self.img_url) if self.img_url else None
+            input = [{ "role": "user", "content": self.user_prompt } ]
+            if img_url:
+                input.append({"type": "input_image", "image_url": f"data:image/png;base64,{img_url}"})
+            schema = dict(model=self.model, tools=tool_schema, input=input, reasoning={ "effort": self.reasoning_effort}, truncation=self.truncation)
+
+        return schema

{versionhq-1.2.4.7 → versionhq-1.2.4.9}/src/versionhq/tool/gpt/file_search.py

@@ -1,3 +1,4 @@
+import datetime
 from typing import List, Dict, Any, Optional, Tuple
 
 from versionhq.tool.gpt import openai_client
@@ -87,6 +88,7 @@ class GPTToolFileSearch:
     max_num_results: int = 2
     include: List[str] = ["output[*].file_search_call.search_results"]
     filters: Optional[FilterSchema] = None
+    _usage: UsageMetrics = UsageMetrics()
 
     def __init__(
             self,
@@ -119,7 +121,8 @@ class GPTToolFileSearch:
     def run(self) -> Tuple[str, List[Dict[str, Any]], UsageMetrics] | None:
         raw_res = ""
         annotations = list()
-        usage = UsageMetrics()
+        usage = self._usage if self._usage else UsageMetrics()
+        start_dt = datetime.datetime.now()
 
         try:
             res = openai_client.responses.create(**self.schema)
@@ -130,10 +133,14 @@ class GPTToolFileSearch:
                 annotations = [{ "index": item.index, "file_id": item.file_id, "filename": item.filename }
                                 for item in res.output[1].content[0].annotations]
                 usage.record_token_usage(**res.usage.__dict__)
-            return raw_res, annotations, usage
+
         except:
             usage.record_errors(ErrorType.TOOL)
-            return raw_res, annotations, usage
+
+        end_dt = datetime.datetime.now()
+        usage.record_latency(start_dt=start_dt, end_dt=end_dt)
+        self._usage = usage
+        return raw_res, annotations, usage
 
 
     @property