freeplay 0.5.0a1__tar.gz → 0.5.0a3__tar.gz

{freeplay-0.5.0a1 → freeplay-0.5.0a3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: freeplay
-Version: 0.5.0a1
+Version: 0.5.0a3
 Summary: 
 License: MIT
 Author: FreePlay Engineering

{freeplay-0.5.0a1 → freeplay-0.5.0a3}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "freeplay"
-version = "0.5.0a1"
+version = "0.5.0a3"
 description = ""
 authors = ["FreePlay Engineering <engineering@freeplay.ai>"]
 license = "MIT"
@@ -17,9 +17,10 @@ pystache = "^0.6.5"
 mypy = "^1"
 types-requests = "^2.31"
 anthropic = { extras = ["bedrock"], version = "^0.39.0" }
-openai = "^1"
+openai = "1.98.0"
 boto3 = "^1.34.97"
-google-cloud-aiplatform = "1.51.0"
+google-cloud-aiplatform = "^1.71.0"
+vertexai = "^1.71.1"
 httpx = "0.27.2"
 
 [tool.poetry.group.test.dependencies]

{freeplay-0.5.0a1 → freeplay-0.5.0a3}/src/freeplay/resources/prompts.py

@@ -8,7 +8,6 @@ from typing import (
     Any,
     Dict,
     List,
-    Literal,
     Optional,
     Protocol,
     Sequence,
@@ -56,7 +55,12 @@ logger = logging.getLogger(__name__)
 class UnsupportedToolSchemaError(FreeplayConfigurationError):
     def __init__(self) -> None:
         super().__init__(
-            f'Tool schema not supported for this model and provider.'
+            'Tool schema not supported for this model and provider.'
+        )
+class VertexAIToolSchemaError(FreeplayConfigurationError):
+    def __init__(self) -> None:
+        super().__init__(
+            'Vertex AI SDK not found. Install google-cloud-aiplatform to get proper Tool objects.'
         )
 
 
@@ -185,6 +189,21 @@ class BoundPrompt:
                     for tool_schema in tool_schema
                 ]
             }
+        elif flavor_name == "gemini_chat":
+            try:
+                from vertexai.generative_models import Tool, FunctionDeclaration  # type: ignore[import-untyped]
+                
+                function_declarations = [
+                    FunctionDeclaration(
+                        name=tool_schema.name,
+                        description=tool_schema.description,
+                        parameters=tool_schema.parameters
+                    )
+                    for tool_schema in tool_schema
+                ]
+                return [Tool(function_declarations=function_declarations)]
+            except ImportError:
+                raise VertexAIToolSchemaError()
 
         raise UnsupportedToolSchemaError()
 

{freeplay-0.5.0a1 → freeplay-0.5.0a3}/src/freeplay/resources/recordings.py

@@ -22,6 +22,7 @@ from freeplay.resources.prompts import (
 )
 from freeplay.resources.sessions import SessionInfo, TraceInfo
 from freeplay.support import CallSupport
+from freeplay.utils import convert_provider_message_to_dict
 
 logger = logging.getLogger(__name__)
 
@@ -129,6 +130,9 @@ class Recordings:
         if len(record_payload.all_messages) < 1:
             raise FreeplayClientError("Messages list must have at least one message. "
                                       "The last message should be the current response.")
+        
+        if record_payload.tool_schema is not None:
+            record_payload.tool_schema = [convert_provider_message_to_dict(tool) for tool in record_payload.tool_schema]
 
         record_api_payload: Dict[str, Any] = {
             "messages": record_payload.all_messages,

{freeplay-0.5.0a1 → freeplay-0.5.0a3}/src/freeplay/utils.py

@@ -75,14 +75,55 @@ def get_user_agent() -> str:
 # Recursively convert Pydantic models, lists, and dicts to dict compatible format -- used to allow us to accept
 # provider message shapes (usually generated types) or the default {'content': ..., 'role': ...} shape.
 def convert_provider_message_to_dict(obj: Any) -> Any:
-    if hasattr(obj, 'model_dump'):
+    """
+    Convert provider message objects to dictionaries.
+    For Vertex AI objects, automatically converts to camelCase.
+    """
+    # List of possible raw attribute names in Vertex AI objects
+    vertex_raw_attrs = [
+        '_raw_content',      # For Content objects
+        '_raw_tool',         # For Tool objects  
+        '_raw_message',      # For message objects
+        '_raw_candidate',    # For Candidate objects
+        '_raw_response',     # For response objects
+        '_raw_function_declaration',  # For FunctionDeclaration
+        '_raw_generation_config',     # For GenerationConfig
+        '_pb',              # Generic protobuf attribute
+    ]
+    
+    # Check for Vertex AI objects with raw protobuf attributes
+    for attr_name in vertex_raw_attrs:
+        if hasattr(obj, attr_name):
+            raw_obj = getattr(obj, attr_name)
+            if raw_obj is not None:
+                try:
+                    # Use the metaclass to_dict with camelCase conversion
+                    return type(raw_obj).to_dict(
+                        raw_obj,
+                        preserving_proto_field_name=False,  # camelCase
+                        use_integers_for_enums=False,  # Keep as strings (we'll lowercase them)
+                        including_default_value_fields=False  # Exclude defaults
+                    )
+                except:  # noqa: E722
+                    # If we can't convert, continue to the next attribute
+                    pass
+    
+    # For non-Vertex AI objects, use their standard to_dict methods
+    if hasattr(obj, 'to_dict') and callable(getattr(obj, 'to_dict')):
+        # Regular to_dict (for Vertex AI wrappers without _raw_* attributes)
+        return obj.to_dict()
+    elif hasattr(obj, 'model_dump'):
         # Pydantic v2
         return obj.model_dump(mode='json')
     elif hasattr(obj, 'dict'):
         # Pydantic v1
         return obj.dict(encode_json=True)
     elif isinstance(obj, dict):
+        # Handle dictionaries recursively
         return {k: convert_provider_message_to_dict(v) for k, v in obj.items()}
     elif isinstance(obj, list):
+        # Handle lists recursively
         return [convert_provider_message_to_dict(item) for item in obj]
+    
+    # Return as-is for primitive types
     return obj