versionhq 1.2.4.13__tar.gz → 1.2.4.15__tar.gz

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/.gitignore

@@ -43,3 +43,5 @@ __pycache__
 
 .env
 .venv/
+
+_screenshots/

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: versionhq
-Version: 1.2.4.13
+Version: 1.2.4.15
 Summary: Autonomous agent networks for task automation with multi-step reasoning.
 Author-email: Kuriko Iwai <kuriko@versi0n.io>
 License: MIT License
@@ -77,6 +77,7 @@ 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"
+Requires-Dist: selenium>=4.30.0; extra == "tools"
 Provides-Extra: torch
 Requires-Dist: torch>=2.6.0; extra == "torch"
 Requires-Dist: torchvision>=0.21.0; extra == "torch"

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/pyproject.toml

@@ -15,7 +15,7 @@ exclude = ["test*", "__pycache__", "*.egg-info"]
 
 [project]
 name = "versionhq"
-version = "1.2.4.13"
+version = "1.2.4.15"
 authors = [{ name = "Kuriko Iwai", email = "kuriko@versi0n.io" }]
 description = "Autonomous agent networks for task automation with multi-step reasoning."
 readme = "README.md"
@@ -82,6 +82,7 @@ tools = [
     "html2text>=2024.2.26",
     "sec-api>=1.0.28",
     "pytest-playwright>=0.7.0",
+    "selenium>=4.30.0",
 ]
 torch = [
     "torch>=2.6.0",

versionhq-1.2.4.15/requirements_tool.txt

@@ -0,0 +1,4 @@
+html2text>=2024.2.26
+sec-api>=1.0.28
+pytest-playwright>=0.7.0
+selenium==4.30.0

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/src/versionhq/__init__.py

@@ -10,7 +10,7 @@ load_dotenv(override=True)
 from versionhq.agent.model import Agent
 from versionhq.agent_network.model import AgentNetwork, Formation, Member, TaskHandlingProcess
 from versionhq.llm.model import LLM
-from versionhq.llm.llm_vars import LLM_CONTEXT_WINDOW_SIZES, MODEL_PARAMS, PROVIDERS, TEXT_MODELS
+from versionhq.llm.llm_vars import LLM_CONTEXT_WINDOW_SIZES, MODEL_PARAMS, PROVIDERS, MODELS
 from versionhq.clients.customer.model import Customer
 from versionhq.clients.product.model import Product, ProductProvider
 from versionhq.clients.workflow.model import MessagingWorkflow, MessagingComponent
@@ -35,7 +35,7 @@ from versionhq.agent_network.formation import form_agent_network
 from versionhq.task_graph.draft import workflow
 
 
-__version__ = "1.2.4.13"
+__version__ = "1.2.4.15"
 __all__ = [
     "Agent",
 
@@ -48,7 +48,7 @@ __all__ = [
     "LLM_CONTEXT_WINDOW_SIZES",
     "MODEL_PARAMS",
     "PROVIDERS",
-    "TEXT_MODELS",
+    "MODELS",
 
     "Customer",
     "Product",

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/src/versionhq/_prompt/model.py

@@ -14,7 +14,6 @@ class Prompt:
     agent: Any = None
     context: Any = None
 
-
     def __init__(self, task, agent, context):
         from versionhq.agent.model import Agent
         from versionhq.task.model import Task
@@ -32,22 +31,24 @@ class Prompt:
         output_prompt = ""
         output_formats_to_follow = dict()
 
-        if self.task.response_schema:
-            if isinstance(self.task.response_schema, list):
-                for item in self.task.response_schema:
-                    if isinstance(item, ResponseField):
-                        output_formats_to_follow[item.title] = f"<Return your answer in {item.data_type.__name__}>"
+        if self.task.is_multimodal == False:
+            if self.task.response_schema:
+                if isinstance(self.task.response_schema, list):
+                    for item in self.task.response_schema:
+                        if isinstance(item, ResponseField):
+                            output_formats_to_follow[item.title] = f"<Return your answer in {item.data_type.__name__}>"
 
-            elif issubclass(self.task.response_schema, BaseModel):
-                for k, v in self.task.response_schema.model_fields.items():
-                    output_formats_to_follow[k] = f"<Return your answer in {v.annotation}>"
+                elif issubclass(self.task.response_schema, BaseModel):
+                    for k, v in self.task.response_schema.model_fields.items():
+                        output_formats_to_follow[k] = f"<Return your answer in {v.annotation}>"
 
-            output_prompt = f"""Your response MUST be a valid JSON string that strictly follows the response format. Use double quotes for all keys and string values. Do not use single quotes, trailing commas, or any other non-standard JSON syntax.
-Ref. Output image: {output_formats_to_follow}
-    """
-        else:
-            output_prompt = "You MUST return your response as a valid JSON serializable string, enclosed in double quotes. Use double quotes for all keys and string values. Do NOT use single quotes, trailing commas, or other non-standard JSON syntax."
+                output_prompt = f"""Your response MUST be a valid JSON string that strictly follows the response format. Use double quotes for all keys and string values. Do not use single quotes, trailing commas, or any other non-standard JSON syntax.
+Ref. Output image: {output_formats_to_follow}"""
+            else:
+                output_prompt = "You MUST return your response as a valid JSON serializable string, enclosed in double quotes. Use double quotes for all keys and string values. Do NOT use single quotes, trailing commas, or other non-standard JSON syntax."
 
+        else:
+            output_prompt = "Return your response in concise manner."
         return dedent(output_prompt)
 
 
@@ -98,19 +99,20 @@ Ref. Output image: {output_formats_to_follow}
 
         content_messages = {}
 
-        if self.task.image:
-            img_url = convert_img_url(self.task.image)
-            if img_url:
-                content_messages.update({ "type": "image_url", "image_url": { "url": img_url }})
+        if self.task.is_multimodal == False:
+            if self.task.image:
+                img_url = convert_img_url(self.task.image)
+                if img_url:
+                    content_messages.update({ "type": "image_url", "image_url": { "url": img_url }})
 
-        if self.task.file:
-            if is_valid_url(self.task.file):
-                content_messages.update({ "type": "image_url", "image_url": self.file })
+            if self.task.file:
+                if is_valid_url(self.task.file):
+                    content_messages.update({ "type": "image_url", "image_url": self.file })
 
-        if self.task.audio and self.agent.llm.provider == "gemini":
-            audio_bytes = Path(self.task.audio).read_bytes()
-            encoded_data = base64.b64encode(audio_bytes).decode("utf-8")
-            content_messages.update({ "type": "image_url", "image_url": "data:audio/mp3;base64,{}".format(encoded_data)})
+            if self.task.audio and self.agent.llm.provider == "gemini":
+                audio_bytes = Path(self.task.audio).read_bytes()
+                encoded_data = base64.b64encode(audio_bytes).decode("utf-8")
+                content_messages.update({ "type": "image_url", "image_url": "data:audio/mp3;base64,{}".format(encoded_data)})
 
         return content_messages
 
@@ -188,7 +190,6 @@ Ref. Output image: {output_formats_to_follow}
         # else:
         #     user_prompt = self.agent._use_trained_data(user_prompt=user_prompt)
 
-
         content_prompt = self._format_content_prompt()
 
         messages = []

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/src/versionhq/_utils/__init__.py

@@ -5,3 +5,4 @@ from versionhq._utils.is_valid_url import is_valid_url
 from versionhq._utils.usage_metrics import UsageMetrics, ErrorType
 from versionhq._utils.convert_img_url import convert_img_url
 from versionhq._utils.is_valid_enum import is_valid_enum
+from versionhq._utils.handle_directory import handle_directory

versionhq-1.2.4.15/src/versionhq/_utils/handle_directory.py

@@ -0,0 +1,15 @@
+import os
+import datetime
+from pathlib import Path
+
+
+def handle_directory(directory_name: str = None, filename: str = None, ext: str = 'png') -> Path:
+    """Creates and returns the absolute file path"""
+
+    os.makedirs(directory_name, exist_ok=True)
+
+    date = str(datetime.datetime.now().strftime('%j'))
+    cwd = Path.cwd()
+    DIRECTORY = cwd / f'{directory_name}/{filename}_{date}.{ext}'
+
+    return DIRECTORY

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/src/versionhq/agent/model.py

@@ -356,12 +356,9 @@ class Agent(BaseModel):
         response_format: Optional[Dict[str, Any]] = None,
         tools: Optional[List[InstanceOf[Tool]| InstanceOf[ToolSet] | Type[Tool]]] = None,
         tool_res_as_final: bool = False,
+        file: str = None, # absolute path to the content file (for multimodal use)
         ) -> Tuple[str, UsageMetrics]:
-        """
-        Create formatted prompts using the developer prompt and the agent's backstory, then call the base model.
-        - Execute the task up to `self.max_retry_limit` times in case of receiving an error or empty response.
-        - Pass the task_tools to the model to let them execute.
-        """
+        """Calls LLM."""
 
         task_execution_counter = 0
         iterations = 0
@@ -375,10 +372,10 @@ class Agent(BaseModel):
             Logger(**self._logger_config, filename=self.key).log(level="info", message=f"Messages sent to the model: {messages}", color="blue")
 
             if tool_res_as_final:
-                raw_response = self.func_calling_llm.call(messages=messages, tools=tools, tool_res_as_final=True)
+                raw_response = self.func_calling_llm.call(messages=messages, tools=tools, tool_res_as_final=True, file=file)
                 usage.record_token_usage(*self.func_calling_llm._usages)
             else:
-                raw_response = self.llm.call(messages=messages, response_format=response_format, tools=tools)
+                raw_response = self.llm.call(messages=messages, response_format=response_format, tools=tools, file=file)
                 usage.record_token_usage(*self.llm._usages)
 
             task_execution_counter += 1
@@ -454,14 +451,14 @@ class Agent(BaseModel):
         return rag_tools, gpt_tools, tools
 
 
-    def _handle_gpt_tools(self, gpt_tools: list[Any] = None) -> Any: # TaskOutput
+    def _handle_gpt_tools(self, gpt_tools: list[Any] = None) -> Any: # TaskOutput or None
         """Generates k, v pairs from multiple GPT tool results and stores them in TaskOutput class."""
 
         from versionhq.task.model import TaskOutput
         from versionhq._utils import UsageMetrics
 
         if not gpt_tools:
-            return
+            return None
 
         tool_res = dict()
         annotation_set = dict()
@@ -470,7 +467,9 @@ class Agent(BaseModel):
         for i, item in enumerate(gpt_tools):
             raw, annotations, usage = item.run()
             tool_res.update({ str(i): raw })
-            annotation_set.update({ str(i): annotations })
+
+            if annotations:
+                annotation_set.update({ str(i): annotations })
             total_usage.aggregate(metrics=usage)
 
         res = TaskOutput(raw=str(tool_res), tool_output=tool_res, usage=total_usage, annotations=annotation_set)
@@ -580,6 +579,7 @@ class Agent(BaseModel):
         raw_response = ""
         user_prompt, dev_prompt = "", ""
         usage = UsageMetrics(id=task.id)
+        file = task.audio if task.is_multimodal and task.audio else task.image if task.is_multimodal and task.image else task.file if task.is_multimodal and task.file else None
 
         if self.max_rpm and self._rpm_controller:
             self._rpm_controller._reset_request_count()
@@ -598,6 +598,7 @@ class Agent(BaseModel):
                 response_format=task._structure_response_format(model_provider=self.llm.provider),
                 tools=tools,
                 tool_res_as_final=task.tool_res_as_final,
+                file=file,
             )
 
         except Exception as e:

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/src/versionhq/llm/llm_vars.py

@@ -28,14 +28,13 @@ PROVIDERS = {
         "HF_ENDPOINT": "HF_ENDPOINT",
     },
     "azure":  {
-        "api_base": "AZURE_OPENAI_ENDPOINT",
+        "api_base": "AZURE_OPENAI_ENDPOINT_MODEL_NAME",
         "api_key": "AZURE_OPENAI_API_KEY",
         "api_version": "AZURE_OPENAI_API_VERSION",
     },
     "azure_ai": {
         "api_key": "AZURE_AI_API_KEY",
         "base_url": "AZURE_AI_API_BASE",
-
     }
 }
 
@@ -47,7 +46,7 @@ ENDPOINTS = [
 
 
 # Resaoning and text generation models
-TEXT_MODELS = {
+MODELS = {
     "openai": [
         "gpt-4.5-preview-2025-02-27",
         "gpt-4",
@@ -96,6 +95,10 @@ TEXT_MODELS = {
         "bedrock/cohere.command-light-text-v14",
     ],
     "azure": [
+        "azure/whisper",
+        "azure/whisper-2",
+        "azure/gpt-4o-mini-audio-preview",
+
         "azure/DeepSeek-V3",
         "azure/DeepSeek-R1",
         "azure/Llama-3.3-70B-Instruct",
@@ -163,6 +166,13 @@ TEXT_MODELS = {
 }
 
 
+AUDIO_TO_TEXT_MODELS = [
+    "azure/whisper",
+    "azure/whisper-2",
+    "azure/gpt-4o-mini-audio-preview",
+]
+
+
 """
 Max input token size by the model.
 """

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/src/versionhq/llm/model.py

@@ -12,9 +12,9 @@ import litellm
 from litellm import JSONSchemaValidationError, get_supported_openai_params, supports_response_schema
 from pydantic import BaseModel, Field, PrivateAttr, model_validator, ConfigDict
 
-from versionhq.llm.llm_vars import LLM_CONTEXT_WINDOW_SIZES, TEXT_MODELS, MODEL_PARAMS, PROVIDERS, ENDPOINTS
+from versionhq.llm.llm_vars import LLM_CONTEXT_WINDOW_SIZES, MODELS, AUDIO_TO_TEXT_MODELS, MODEL_PARAMS, PROVIDERS, ENDPOINTS
 from versionhq.tool.model import Tool, ToolSet
-from versionhq._utils import Logger
+from versionhq._utils import Logger, UsageMetrics, ErrorType
 
 
 load_dotenv(override=True)
@@ -115,7 +115,7 @@ class LLM(BaseModel):
                 self.provider = DEFAULT_MODEL_PROVIDER_NAME
 
             else:
-                provider_model_list = TEXT_MODELS.get(self.provider)
+                provider_model_list = MODELS.get(self.provider)
                 if provider_model_list:
                     self.model = provider_model_list[0]
                     self.provider = self.provider
@@ -127,29 +127,29 @@ class LLM(BaseModel):
         elif self.model and self.provider is None:
             model_match = [
                 item for item in [
-                    [val for val in v if val == self.model][0] for k, v in TEXT_MODELS.items() if [val for val in v if val == self.model]
+                    [val for val in v if val == self.model][0] for k, v in MODELS.items() if [val for val in v if val == self.model]
                 ] if item
             ]
             model_partial_match = [
                 item for item in [
-                    [val for val in v if val.find(self.model) != -1][0] for k, v in TEXT_MODELS.items() if [val for val in v if val.find(self.model) != -1]
+                    [val for val in v if val.find(self.model) != -1][0] for k, v in MODELS.items() if [val for val in v if val.find(self.model) != -1]
                 ] if item
             ]
-            provider_match = [k for k, v in TEXT_MODELS.items() if k == self.model]
+            provider_match = [k for k, v in MODELS.items() if k == self.model]
 
             if model_match:
                 self.model = model_match[0]
-                self.provider = [k for k, v in TEXT_MODELS.items() if self.model in v][0]
+                self.provider = [k for k, v in MODELS.items() if self.model in v][0]
 
             elif model_partial_match:
                 self.model = model_partial_match[0]
-                self.provider = [k for k, v in TEXT_MODELS.items() if [item for item in v if item.find(self.model) != -1]][0]
+                self.provider = [k for k, v in MODELS.items() if [item for item in v if item.find(self.model) != -1]][0]
 
             elif provider_match:
                 provider = provider_match[0]
-                if self.TEXT_MODELS.get(provider):
+                if self.MODELS.get(provider):
                     self.provider = provider
-                    self.model = self.TEXT_MODELS.get(provider)[0]
+                    self.model = self.MODELS.get(provider)[0]
                 else:
                     self.provider = DEFAULT_MODEL_PROVIDER_NAME
                     self.model = DEFAULT_MODEL_NAME
@@ -159,7 +159,7 @@ class LLM(BaseModel):
                 self.provider = DEFAULT_MODEL_PROVIDER_NAME
 
         else:
-            provider_model_list = TEXT_MODELS.get(self.provider)
+            provider_model_list = MODELS.get(self.provider)
             if self.model not in provider_model_list:
                 self._logger.log(level="warning", message=f"The provided model: {self._init_model_name} is not in the list. We will assign a default model.", color="yellow")
                 self.model = DEFAULT_MODEL_NAME
@@ -232,7 +232,16 @@ class LLM(BaseModel):
 
         valid_cred = {}
         for k, v in cred.items():
-            val = os.environ.get(v, None)
+            val = None
+            if '_MODEL_NAME' in v:
+                model_name = self.model.split('/')[-1] if self.model.split('/') else self.model
+                key = v.replace('_MODEL_NAME', f'_{model_name.replace("-", '_').replace(' ', '_').upper()}')
+                val = os.environ.get(key, None)
+                if not val:
+                    val = os.environ.get(v.replace('_MODEL_NAME', ''), None)
+            else:
+                val = os.environ.get(v, None)
+
             if val:
                 valid_cred[str(k)] = val
 
@@ -288,12 +297,12 @@ class LLM(BaseModel):
         messages: List[Dict[str, str]],
         response_format: Optional[Dict[str, Any]] = None,
         tools: Optional[List[Tool | ToolSet | Any ]] = None,
-        config: Optional[Dict[str, Any]] = {}, # any other conditions to pass on to the model.
-        tool_res_as_final: bool = False
+        config: Optional[Dict[str, Any]] = dict(),
+        tool_res_as_final: bool = False,
+        file: str = None
     ) -> str:
-        """
-        Execute LLM based on the agent's params and model params.
-        """
+        """Configures and calls the LLM (chat, text generation, reasoning models)."""
+
         litellm.drop_params = True
         litellm.set_verbose = True
 
@@ -302,9 +311,32 @@ class LLM(BaseModel):
                 self._set_callbacks(self.callbacks)
 
             try:
-                res, tool_res = None, ""
+                res = None
+                tool_res = ""
                 cred = self._set_credentials()
 
+                if file and self.model in AUDIO_TO_TEXT_MODELS:
+                    params = self._create_valid_params(config=config)
+                    audio_file = open(file, 'rb')
+                    res = litellm.transcription(
+                        model=self.model,
+                        file=audio_file,
+                        rompt=messages,
+                        ustom_llm_provider=self.endpoint_provider,
+                        response_format="json",
+                        **cred
+                    )
+                    usage = UsageMetrics()
+                    if res:
+                        usage.latency = res._response_ms if hasattr(res, '_response_ms') else 0
+                        self._usages.append(usage)
+                        return res.text
+                    else:
+                        usage.record_errors(type=ErrorType.API)
+                        self._usages.append(usage)
+                        return None
+
+
                 if self.provider == "gemini":
                     self.response_format = { "type": "json_object" } if not tools and self.model != "gemini/gemini-2.0-flash-thinking-exp" else None
                 elif response_format and "json_schema" in response_format:

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/src/versionhq/task/model.py

@@ -314,6 +314,7 @@ class Task(BaseModel):
     name: Optional[str] = Field(default=None)
     description: str = Field(description="Description of the actual task")
     response_schema: Optional[Type[BaseModel] | List[ResponseField]] = Field(default=None, description="stores response format")
+    is_multimodal: bool = False
 
     # tool usage
     tools: Optional[List[Any]] = Field(default_factory=list, description="tools that the agent can use aside from their tools")

{versionhq-1.2.4.13 → versionhq-1.2.4.15}/src/versionhq/tool/gpt/_enum.py

@@ -7,7 +7,7 @@ class GPTSizeEnum(str, Enum):
     HIGH = "high"
 
 
-class GPTCUAEnvironmentEnum(str, Enum):
+class GPTCUABrowserEnum(str, Enum):
     BROWSER = "browser"
     MAC = "mac"
     WINDOWS = "windows"