payi 0.1.0a72__tar.gz → 0.1.0a73__tar.gz

payi-0.1.0a73/.release-please-manifest.json

@@ -0,0 +1,3 @@
+{
+  ".": "0.1.0-alpha.73"
+}

{payi-0.1.0a72 → payi-0.1.0a73}/CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## 0.1.0-alpha.73 (2025-05-13)
+
+Full Changelog: [v0.1.0-alpha.72...v0.1.0-alpha.73](https://github.com/Pay-i/pay-i-python/compare/v0.1.0-alpha.72...v0.1.0-alpha.73)
+
+### Features
+
+* Anthropic Vertex, OpenAi Responses ([#297](https://github.com/Pay-i/pay-i-python/issues/297)) ([535163f](https://github.com/Pay-i/pay-i-python/commit/535163f1d29f39838a16dc87a4ef40542b5d93b0))
+
 ## 0.1.0-alpha.72 (2025-05-10)
 
 Full Changelog: [v0.1.0-alpha.71...v0.1.0-alpha.72](https://github.com/Pay-i/pay-i-python/compare/v0.1.0-alpha.71...v0.1.0-alpha.72)

{payi-0.1.0a72 → payi-0.1.0a73}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: payi
-Version: 0.1.0a72
+Version: 0.1.0a73
 Summary: The official Python library for the payi API
 Project-URL: Homepage, https://github.com/Pay-i/pay-i-python
 Project-URL: Repository, https://github.com/Pay-i/pay-i-python

{payi-0.1.0a72 → payi-0.1.0a73}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "payi"
-version = "0.1.0-alpha.72"
+version = "0.1.0-alpha.73"
 description = "The official Python library for the payi API"
 dynamic = ["readme"]
 license = "Apache-2.0"

{payi-0.1.0a72 → payi-0.1.0a73}/src/payi/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "payi"
-__version__ = "0.1.0-alpha.72"  # x-release-please-version
+__version__ = "0.1.0-alpha.73"  # x-release-please-version

{payi-0.1.0a72 → payi-0.1.0a73}/src/payi/lib/AnthropicInstrumentor.py

@@ -11,7 +11,13 @@ from payi.types.ingest_units_params import Units
 from .instrument import _IsStreaming, _ProviderRequest, _PayiInstrumentor
 
 
-class AnthropicIntrumentor:
+class AnthropicInstrumentor:
+    @staticmethod
+    def is_vertex(instance: Any) -> bool:
+        from anthropic import AnthropicVertex, AsyncAnthropicVertex  # type: ignore # noqa: I001
+
+        return isinstance(instance._client, (AsyncAnthropicVertex, AnthropicVertex))
+
     @staticmethod
     def instrument(instrumentor: _PayiInstrumentor) -> None:
         try:
@@ -20,25 +26,25 @@ class AnthropicIntrumentor:
             wrap_function_wrapper(
                 "anthropic.resources.messages",
                 "Messages.create",
-                chat_wrapper(instrumentor),
+                messages_wrapper(instrumentor),
             )
 
             wrap_function_wrapper(
                 "anthropic.resources.messages",
                 "Messages.stream",
-                chat_wrapper(instrumentor),
+                messages_wrapper(instrumentor),
             )
 
             wrap_function_wrapper(
                 "anthropic.resources.messages",
                 "AsyncMessages.create",
-                achat_wrapper(instrumentor),
+                amessages_wrapper(instrumentor),
             )
 
             wrap_function_wrapper(
                 "anthropic.resources.messages",
                 "AsyncMessages.stream",
-                achat_wrapper(instrumentor),
+                amessages_wrapper(instrumentor),
             )
 
         except Exception as e:
@@ -47,7 +53,7 @@ class AnthropicIntrumentor:
 
 
 @_PayiInstrumentor.payi_wrapper
-def chat_wrapper(
+def messages_wrapper(
     instrumentor: _PayiInstrumentor,
     wrapped: Any,
     instance: Any,
@@ -55,7 +61,7 @@ def chat_wrapper(
     **kwargs: Any,
 ) -> Any:
     return instrumentor.invoke_wrapper(
-        _AnthropicProviderRequest(instrumentor),
+        _AnthropicProviderRequest(instrumentor, instance),
         _IsStreaming.kwargs,
         wrapped,
         instance,
@@ -64,7 +70,7 @@ def chat_wrapper(
     )
 
 @_PayiInstrumentor.payi_awrapper
-async def achat_wrapper(
+async def amessages_wrapper(
     instrumentor: _PayiInstrumentor,
     wrapped: Any,
     instance: Any,
@@ -72,7 +78,7 @@ async def achat_wrapper(
     **kwargs: Any,
 ) -> Any:
     return await instrumentor.async_invoke_wrapper(
-        _AnthropicProviderRequest(instrumentor),
+        _AnthropicProviderRequest(instrumentor, instance),
         _IsStreaming.kwargs,
         wrapped,
         instance,
@@ -81,8 +87,12 @@ async def achat_wrapper(
     )
 
 class _AnthropicProviderRequest(_ProviderRequest):
-    def __init__(self, instrumentor: _PayiInstrumentor):
-        super().__init__(instrumentor=instrumentor, category=PayiCategories.anthropic)
+    def __init__(self, instrumentor: _PayiInstrumentor, instance: Any = None) -> None:
+        self._vertex: bool = AnthropicInstrumentor.is_vertex(instance)
+        super().__init__(
+            instrumentor=instrumentor,
+            category=PayiCategories.google_vertex if self._vertex else PayiCategories.anthropic
+            )
 
     @override
     def process_chunk(self, chunk: Any) -> bool:
@@ -138,7 +148,8 @@ class _AnthropicProviderRequest(_ProviderRequest):
 
     @override
     def process_request(self, instance: Any, extra_headers: 'dict[str, str]', args: Sequence[Any], kwargs: Any) -> bool:
-        self._ingest["resource"] = kwargs.get("model", "")
+        self._ingest["resource"] = ("anthropic." if self._vertex else "") + kwargs.get("model", "")
+
         messages = kwargs.get("messages")
         if messages:
             estimated_token_count = 0 

{payi-0.1.0a72 → payi-0.1.0a73}/src/payi/lib/OpenAIInstrumentor.py

@@ -7,7 +7,6 @@ from importlib.metadata import version
 import tiktoken  # type: ignore
 from wrapt import wrap_function_wrapper  # type: ignore
 
-from payi.types import IngestUnitsParams
 from payi.lib.helpers import PayiCategories, PayiHeaderNames
 from payi.types.ingest_units_params import Units
 
@@ -50,6 +49,18 @@ class OpenAiInstrumentor:
                 aembeddings_wrapper(instrumentor),
             )
 
+            wrap_function_wrapper(
+                "openai.resources.responses",
+                "Responses.create",
+                responses_wrapper(instrumentor),
+            )
+
+            wrap_function_wrapper(
+                "openai.resources.responses",
+                "AsyncResponses.create",
+                aresponses_wrapper(instrumentor),
+            )
+
         except Exception as e:
             logging.debug(f"Error instrumenting openai: {e}")
             return
@@ -122,10 +133,55 @@ async def achat_wrapper(
         args,
         kwargs,
     )
+    
+@_PayiInstrumentor.payi_wrapper
+def responses_wrapper(
+    instrumentor: _PayiInstrumentor,
+    wrapped: Any,
+    instance: Any,
+    *args: Any,
+    **kwargs: Any,
+) -> Any:
+    return instrumentor.invoke_wrapper(
+        _OpenAiResponsesProviderRequest(instrumentor),
+        _IsStreaming.kwargs,
+        wrapped,
+        instance,
+        args,
+        kwargs,
+    )
+
+@_PayiInstrumentor.payi_awrapper
+async def aresponses_wrapper(
+    instrumentor: _PayiInstrumentor,
+    wrapped: Any,
+    instance: Any,
+    *args: Any,
+    **kwargs: Any,
+) -> Any:
+    return await instrumentor.async_invoke_wrapper(
+        _OpenAiResponsesProviderRequest(instrumentor),
+        _IsStreaming.kwargs,
+        wrapped,
+        instance,
+        args,
+        kwargs,
+    )
 
 class _OpenAiProviderRequest(_ProviderRequest):
-    def __init__(self, instrumentor: _PayiInstrumentor):
+    chat_input_tokens_key: str = "prompt_tokens"
+    chat_output_tokens_key: str = "completion_tokens"
+    chat_input_tokens_details_key: str = "prompt_tokens_details"
+
+    responses_input_tokens_key: str = "input_tokens"
+    responses_output_tokens_key: str = "output_tokens"
+    responses_input_tokens_details_key: str = "input_tokens_details"
+
+    def __init__(self, instrumentor: _PayiInstrumentor, input_tokens_key: str, output_tokens_key: str, input_tokens_details_key: str) -> None:
         super().__init__(instrumentor=instrumentor, category=PayiCategories.openai)
+        self._input_tokens_key = input_tokens_key
+        self._output_tokens_key = output_tokens_key
+        self._input_tokens_details_key = input_tokens_details_key
 
     @override
     def process_request(self, instance: Any, extra_headers: 'dict[str, str]', args: Sequence[Any], kwargs: Any) -> bool: # type: ignore
@@ -135,16 +191,19 @@ class _OpenAiProviderRequest(_ProviderRequest):
             return True
 
         context = self._instrumentor.get_context_safe()
-        route_as_resource = extra_headers.get(PayiHeaderNames.route_as_resource) or context.get("route_as_resource")
+        price_as_category = extra_headers.get(PayiHeaderNames.price_as_category) or context.get("price_as_category")
+        price_as_resource = extra_headers.get(PayiHeaderNames.price_as_resource) or context.get("price_as_resource")
         resource_scope = extra_headers.get(PayiHeaderNames.resource_scope) or context.get("resource_scope")
 
-        if PayiHeaderNames.route_as_resource in extra_headers:
-            del extra_headers[PayiHeaderNames.route_as_resource]
+        if PayiHeaderNames.price_as_category in extra_headers:
+            del extra_headers[PayiHeaderNames.price_as_category]
+        if PayiHeaderNames.price_as_resource in extra_headers:
+            del extra_headers[PayiHeaderNames.price_as_resource]
         if PayiHeaderNames.resource_scope in extra_headers:
             del extra_headers[PayiHeaderNames.resource_scope]
             
-        if not route_as_resource:
-            logging.error("Azure OpenAI route as resource not found, not ingesting")
+        if not price_as_resource and not price_as_category:
+            logging.error("Azure OpenAI requires price as resource and/or category to be specified, not ingesting")
             return False
 
         if resource_scope:
@@ -157,8 +216,13 @@ class _OpenAiProviderRequest(_ProviderRequest):
         self._category = PayiCategories.azure_openai
 
         self._ingest["category"] = self._category
-        self._ingest["resource"] = route_as_resource
- 
+
+        if price_as_category:
+            # price as category overrides default
+            self._ingest["category"] = price_as_category
+        if price_as_resource:
+            self._ingest["resource"] = price_as_resource
+
         return True
 
     @override
@@ -193,9 +257,59 @@ class _OpenAiProviderRequest(_ProviderRequest):
 
         return True
 
+    def process_synchronous_response_worker(
+        self,
+        response: str,
+        log_prompt_and_response: bool,
+        ) -> Any:
+        response_dict = model_to_dict(response)
+
+        self.add_usage_units(response_dict.get("usage", {}))
+
+        if log_prompt_and_response:
+            self._ingest["provider_response_json"] = [json.dumps(response_dict)]
+
+        if "id" in response_dict:
+            self._ingest["provider_response_id"] = response_dict["id"]
+
+        return None
+
+    def add_usage_units(self, usage: "dict[str, Any]",) -> None:
+        units = self._ingest["units"]
+
+        input = usage[self._input_tokens_key] if self._input_tokens_key in usage else 0
+        output = usage[self._output_tokens_key] if self._output_tokens_key in usage else 0
+        input_cache = 0
+
+        prompt_tokens_details = usage.get(self._input_tokens_details_key)
+        if prompt_tokens_details:
+            input_cache = prompt_tokens_details.get("cached_tokens", 0)
+            if input_cache != 0:
+                units["text_cache_read"] = Units(input=input_cache, output=0)
+
+        input = _PayiInstrumentor.update_for_vision(input - input_cache, units, self._estimated_prompt_tokens)
+
+        units["text"] = Units(input=input, output=output)
+
+    @staticmethod
+    def has_image_and_get_texts(encoding: tiktoken.Encoding, content: Union[str, 'list[Any]'], image_type: str = "image_url", text_type:str = "text") -> 'tuple[bool, int]':
+        if isinstance(content, str):
+            return False, 0
+        elif isinstance(content, list): # type: ignore
+            has_image = any(item.get("type") == image_type for item in content)
+            if has_image is False:
+                return has_image, 0
+            
+            token_count = sum(len(encoding.encode(item.get("text", ""))) for item in content if item.get("type") == text_type)
+            return has_image, token_count
+
 class _OpenAiEmbeddingsProviderRequest(_OpenAiProviderRequest):
     def __init__(self, instrumentor: _PayiInstrumentor):
-        super().__init__(instrumentor=instrumentor)
+        super().__init__(
+            instrumentor=instrumentor,
+            input_tokens_key=_OpenAiProviderRequest.chat_input_tokens_key,
+            output_tokens_key=_OpenAiProviderRequest.chat_output_tokens_key,
+            input_tokens_details_key=_OpenAiProviderRequest.chat_input_tokens_details_key)
 
     @override
     def process_synchronous_response(
@@ -203,11 +317,16 @@ class _OpenAiEmbeddingsProviderRequest(_OpenAiProviderRequest):
         response: Any,
         log_prompt_and_response: bool,
         kwargs: Any) -> Any:
-        return process_chat_synchronous_response(response, self._ingest, log_prompt_and_response, self._estimated_prompt_tokens)
+        return self.process_synchronous_response_worker(response, log_prompt_and_response)
 
 class _OpenAiChatProviderRequest(_OpenAiProviderRequest):
     def __init__(self, instrumentor: _PayiInstrumentor):
-        super().__init__(instrumentor=instrumentor)
+        super().__init__(
+            instrumentor=instrumentor,
+            input_tokens_key=_OpenAiProviderRequest.chat_input_tokens_key,
+            output_tokens_key=_OpenAiProviderRequest.chat_output_tokens_key,
+            input_tokens_details_key=_OpenAiProviderRequest.chat_input_tokens_details_key)
+
         self._include_usage_added = False
 
     @override
@@ -223,9 +342,9 @@ class _OpenAiChatProviderRequest(_OpenAiProviderRequest):
 
         usage = model.get("usage")
         if usage:
-            add_usage_units(usage, self._ingest["units"], self._estimated_prompt_tokens)
+            self.add_usage_units(usage)
 
-            # If we aded "include_usage" in the request on behalf of the client, do not return the extra 
+            # If we added "include_usage" in the request on behalf of the client, do not return the extra 
             # packet which contains the usage to the client as they are not expecting the data
             if self._include_usage_added:
                 send_chunk_to_client = False
@@ -249,7 +368,7 @@ class _OpenAiChatProviderRequest(_OpenAiProviderRequest):
                 enc = tiktoken.get_encoding("o200k_base") # type: ignore
             
             for message in messages:
-                msg_has_image, msg_prompt_tokens = has_image_and_get_texts(enc, message.get('content', ''))
+                msg_has_image, msg_prompt_tokens = self.has_image_and_get_texts(enc, message.get('content', ''))
                 if msg_has_image:
                     has_image = True
                     estimated_token_count += msg_prompt_tokens
@@ -276,20 +395,86 @@ class _OpenAiChatProviderRequest(_OpenAiProviderRequest):
         response: Any,
         log_prompt_and_response: bool,
         kwargs: Any) -> Any:
-        process_chat_synchronous_response(response, self._ingest, log_prompt_and_response, self._estimated_prompt_tokens)
+        return self.process_synchronous_response_worker(response, log_prompt_and_response)
+
+class _OpenAiResponsesProviderRequest(_OpenAiProviderRequest):
+    def __init__(self, instrumentor: _PayiInstrumentor):
+        super().__init__(
+            instrumentor=instrumentor,
+            input_tokens_key=_OpenAiProviderRequest.responses_input_tokens_key,
+            output_tokens_key=_OpenAiProviderRequest.responses_output_tokens_key,
+            input_tokens_details_key=_OpenAiProviderRequest.responses_input_tokens_details_key)
 
-def process_chat_synchronous_response(response: str, ingest: IngestUnitsParams, log_prompt_and_response: bool, estimated_prompt_tokens: Optional[int]) -> Any:
-    response_dict = model_to_dict(response)
+    @override
+    def process_chunk(self, chunk: Any) -> bool:
+        model = model_to_dict(chunk)
+        response: dict[str, Any] = model.get("response", {})
+
+        if "provider_response_id" not in self._ingest:
+            response_id = response.get("id", None)
+            if response_id:
+                self._ingest["provider_response_id"] = response_id
+
+        usage = response.get("usage")
+        if usage:
+            self.add_usage_units(usage)
+
+        return True
 
-    add_usage_units(response_dict.get("usage", {}), ingest["units"], estimated_prompt_tokens)
+    @override
+    def process_request(self, instance: Any, extra_headers: 'dict[str, str]', args: Sequence[Any], kwargs: Any) -> bool:
+        result = super().process_request(instance, extra_headers, args, kwargs)
+        if result is False:
+            return result
+        
+        input: ResponseInputParam  = kwargs.get("input", None) # type: ignore
+        if not input or isinstance(input, str) or not isinstance(input, list):
+            return True
+        
+        estimated_token_count = 0 
+        has_image = False
 
-    if log_prompt_and_response:
-        ingest["provider_response_json"] = [json.dumps(response_dict)]
+        try: 
+            enc = tiktoken.encoding_for_model(kwargs.get("model")) # type: ignore
+        except KeyError:
+            enc = tiktoken.get_encoding("o200k_base") # type: ignore
+        
+        # find each content..type="input_text" and count tokens
+        # input=[{
+        #     "role": "user",
+        #     "content": [
+        #         {
+        #             "type": "input_text",
+        #             "text": "what's in this image?"
+        #         },
+        #         {
+        #             "type": "input_image",
+        #             "image_url": ... 
+        #         },
+        #     ],
+        # }]
+        for item in input: # type: ignore
+            if isinstance(item, dict):
+                for key, value in item.items(): # type: ignore
+                    if key == "content":
+                        if isinstance(value, list):
+                            msg_has_image, msg_prompt_tokens = self.has_image_and_get_texts(enc, value, image_type="input_image", text_type="input_text") # type: ignore 
+                            if msg_has_image:
+                                has_image = True
+                                estimated_token_count += msg_prompt_tokens
+
+        if has_image and estimated_token_count > 0:
+            self._estimated_prompt_tokens = estimated_token_count
 
-    if "id" in response_dict:
-        ingest["provider_response_id"] = response_dict["id"]
+        return True
 
-    return None
+    @override
+    def process_synchronous_response(
+        self,
+        response: Any,
+        log_prompt_and_response: bool,
+        kwargs: Any) -> Any:
+        return self.process_synchronous_response_worker(response, log_prompt_and_response)
 
 def model_to_dict(model: Any) -> Any:
     if version("pydantic") < "2.0.0":
@@ -299,31 +484,4 @@ def model_to_dict(model: Any) -> Any:
     elif hasattr(model, "parse"):  # Raw API response
         return model_to_dict(model.parse())
     else:
-        return model
-
-
-def add_usage_units(usage: "dict[str, Any]", units: "dict[str, Units]", estimated_prompt_tokens: Optional[int]) -> None:
-    input = usage["prompt_tokens"] if "prompt_tokens" in usage else 0
-    output = usage["completion_tokens"] if "completion_tokens" in usage else 0
-    input_cache = 0
-
-    prompt_tokens_details = usage.get("prompt_tokens_details")
-    if prompt_tokens_details:
-        input_cache = prompt_tokens_details.get("cached_tokens", 0)
-        if input_cache != 0:
-            units["text_cache_read"] = Units(input=input_cache, output=0)
-
-    input = _PayiInstrumentor.update_for_vision(input - input_cache, units, estimated_prompt_tokens)
-
-    units["text"] = Units(input=input, output=output)
-
-def has_image_and_get_texts(encoding: tiktoken.Encoding, content: Union[str, 'list[Any]']) -> 'tuple[bool, int]':
-    if isinstance(content, str):
-        return False, 0
-    elif isinstance(content, list): # type: ignore
-        has_image = any(item.get("type") == "image_url" for item in content)
-        if has_image is False:
-            return has_image, 0
-        
-        token_count = sum(len(encoding.encode(item.get("text", ""))) for item in content if item.get("type") == "text")
-        return has_image, token_count
+        return model        

{payi-0.1.0a72 → payi-0.1.0a73}/src/payi/lib/helpers.py

@@ -11,8 +11,10 @@ class PayiHeaderNames:
     use_case_id:str = "xProxy-UseCase-ID"
     use_case_name:str = "xProxy-UseCase-Name"
     use_case_version:str = "xProxy-UseCase-Version"
+    use_case_step:str = "xProxy-UseCase-Step"
     user_id:str = "xProxy-User-ID"
-    route_as_resource:str = "xProxy-RouteAs-Resource"
+    price_as_category:str = "xProxy-PriceAs-Category"
+    price_as_resource:str = "xProxy-PriceAs-Resource"
     provider_base_uri = "xProxy-Provider-BaseUri"
     resource_scope:str = "xProxy-Resource-Scope"
     api_key:str = "xProxy-Api-Key"
@@ -32,7 +34,6 @@ def create_limit_header_from_ids(limit_ids: List[str]) -> Dict[str, str]:
 
     return { PayiHeaderNames.limit_ids: ",".join(valid_ids) } if valid_ids else {}
 
-
 def create_request_header_from_tags(request_tags: List[str]) -> Dict[str, str]:
     if not isinstance(request_tags, list):  # type: ignore
         raise TypeError("request_tags must be a list")
@@ -50,7 +51,9 @@ def create_headers(
     use_case_id: Union[str, None] = None,
     use_case_name: Union[str, None] = None,
     use_case_version: Union[int, None] = None,
-    route_as_resource: Union[str, None] = None,
+    use_case_step: Union[str, None] = None,
+    price_as_category: Union[str, None] = None,
+    price_as_resource: Union[str, None] = None,
     resource_scope: Union[str, None] = None,
 ) -> Dict[str, str]:
     headers: Dict[str, str] = {}
@@ -71,8 +74,12 @@ def create_headers(
         headers.update({ PayiHeaderNames.use_case_name: use_case_name})
     if use_case_version:
         headers.update({ PayiHeaderNames.use_case_version: str(use_case_version)})
-    if route_as_resource:
-        headers.update({ PayiHeaderNames.route_as_resource: route_as_resource})
+    if use_case_step:
+        headers.update({ PayiHeaderNames.use_case_step: use_case_step})
+    if price_as_category:
+        headers.update({ PayiHeaderNames.price_as_category: price_as_category})
+    if price_as_resource:
+        headers.update({ PayiHeaderNames.price_as_resource: price_as_resource})
     if resource_scope:
         headers.update({ PayiHeaderNames.resource_scope: resource_scope })
     return headers
@@ -99,3 +106,6 @@ def payi_azure_openai_url(payi_base_url: Union[str, None] = None) -> str:
 
 def payi_aws_bedrock_url(payi_base_url: Union[str, None] = None) -> str:
     return _resolve_payi_base_url(payi_base_url=payi_base_url) + "/api/v1/proxy/aws.bedrock"
+
+# def payi_google_vertex_url(payi_base_url: Union[str, None] = None) -> str:
+#     return _resolve_payi_base_url(payi_base_url=payi_base_url) + "/api/v1/proxy/google.vertex"