paygent-sdk 2.0.0__tar.gz → 3.0.0__tar.gz

{paygent_sdk-2.0.0/paygent_sdk.egg-info → paygent_sdk-3.0.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: paygent-sdk
-Version: 2.0.0
+Version: 3.0.0
 Summary: Official Python SDK for Paygent - Track AI usage and costs across multiple providers (OpenAI, Anthropic, Google, DeepSeek, etc.)
 Home-page: https://github.com/paygent/paygent-sdk-python
 Author: Paygent

{paygent_sdk-2.0.0 → paygent_sdk-3.0.0}/paygent_sdk/__init__.py

@@ -13,24 +13,11 @@ from .models import (
 )
 from .voice_client import send_stt_usage, send_tts_usage  # Import to attach methods to Client
 
-# Import wrappers with optional LangChain support
-try:
-    from .wrappers import (
-        PaygentOpenAI,
-        PaygentAnthropic,
-        PaygentMistral,
-        PaygentGemini,
-        PaygentLangChainCallback
-    )
-    _has_langchain = True
-except ImportError:
-    from .wrappers import (
-        PaygentOpenAI,
-        PaygentAnthropic,
-        PaygentMistral,
-        PaygentGemini
-    )
-    _has_langchain = False
+# Wrappers are lazily imported in the wrappers module to avoid requiring
+# installation of peer dependencies (openai, anthropic, mistral, etc.) that aren't being used.
+# You can still import them normally:
+#   from paygent_sdk import PaygentOpenAI, PaygentGemini, etc.
+# But they will only actually load when first accessed.
 from .constants import (
     ServiceProvider,
     OpenAIModels,
@@ -41,14 +28,18 @@ from .constants import (
     MistralAIModels,
     CohereModels,
     DeepSeekModels,
+    MoonshotAIModels,
     DeepgramSTTModels,
     MicrosoftAzureSpeechSTTModels,
     GoogleCloudSpeechSTTModels,
     AssemblyAISTTModels,
+    ElevenLabsSTTModels,
+    SonioxSTTModels,
     AmazonPollyTTSModels,
     MicrosoftAzureSpeechTTSModels,
     GoogleCloudTextToSpeechTTSModels,
     DeepgramTTSModels,
+    ElevenLabsTTSModels,
     is_model_supported
 )
 
@@ -84,17 +75,38 @@ __all__ = [
     "MistralAIModels",
     "CohereModels",
     "DeepSeekModels",
+    "MoonshotAIModels",
     
     # STT/TTS Model constants
     "DeepgramSTTModels",
     "MicrosoftAzureSpeechSTTModels",
     "GoogleCloudSpeechSTTModels",
     "AssemblyAISTTModels",
+    "ElevenLabsSTTModels",
+    "SonioxSTTModels",
     "AmazonPollyTTSModels",
     "MicrosoftAzureSpeechTTSModels",
     "GoogleCloudTextToSpeechTTSModels",
     "DeepgramTTSModels",
+    "ElevenLabsTTSModels",
     
     # Utility functions
     "is_model_supported"
 ]
+
+
+def __getattr__(name):
+    """
+    Lazy import wrapper classes to avoid requiring peer dependencies that aren't being used.
+    
+    This allows importing wrappers like:
+        from paygent_sdk import PaygentOpenAI
+    
+    But the actual import only happens when accessed, so if you never use PaygentOpenAI,
+    you don't need the openai package installed.
+    """
+    if name in ["PaygentOpenAI", "PaygentAnthropic", "PaygentMistral", "PaygentGemini", "PaygentLangChainCallback"]:
+        from . import wrappers
+        return getattr(wrappers, name)
+    
+    raise AttributeError(f"module '{__name__}' has no attribute '{name}'")

{paygent_sdk-2.0.0 → paygent_sdk-3.0.0}/paygent_sdk/client.py

@@ -101,12 +101,25 @@ class Client:
 
         # Calculate cost per 1000 tokens
         prompt_cost = (usage_data.prompt_tokens / 1000.0) * pricing.prompt_tokens_cost
+        
+        # Handle cached tokens: if model doesn't support caching (cached_tokens_cost is None),
+        # bill cached tokens at regular prompt token rate
+        cached_cost = 0.0
+        if usage_data.cached_tokens and usage_data.cached_tokens > 0:
+            if pricing.cached_tokens_cost is not None:
+                # Model supports caching - use cached token price
+                cached_cost = (usage_data.cached_tokens / 1000.0) * pricing.cached_tokens_cost
+            else:
+                # Model doesn't support caching - bill at prompt token rate
+                cached_cost = (usage_data.cached_tokens / 1000.0) * pricing.prompt_tokens_cost
+        
         completion_cost = (usage_data.completion_tokens / 1000.0) * pricing.completion_tokens_cost
-        total_cost = prompt_cost + completion_cost
+        total_cost = prompt_cost + cached_cost + completion_cost
 
         self.logger.debug(
             f"Cost calculation for model '{model}': "
             f"prompt_tokens={usage_data.prompt_tokens} ({prompt_cost:.6f}), "
+            f"cached_tokens={usage_data.cached_tokens or 0} ({cached_cost:.6f}), "
             f"completion_tokens={usage_data.completion_tokens} ({completion_cost:.6f}), "
             f"total={total_cost:.6f}"
         )
@@ -135,9 +148,25 @@ class Client:
         """
         # Removed verbose logging - only log errors
 
+        # 🎯 AUTOMATIC CACHED TOKEN HANDLING
+        # Users can pass total prompt tokens - we automatically subtract cached tokens
+        # This makes manual tracking easier (no math required!)
+        cached_tokens = usage_data.cached_tokens or 0
+        regular_prompt_tokens = usage_data.prompt_tokens - cached_tokens
+
+        # Calculate cost using separated token counts
+        adjusted_usage_data = UsageData(
+            service_provider=usage_data.service_provider,
+            model=usage_data.model,
+            prompt_tokens=regular_prompt_tokens,
+            completion_tokens=usage_data.completion_tokens,
+            total_tokens=usage_data.total_tokens,
+            cached_tokens=cached_tokens
+        )
+
         # Calculate cost
         try:
-            cost = self._calculate_cost(usage_data.model, usage_data)
+            cost = self._calculate_cost(usage_data.model, adjusted_usage_data)
         except Exception as e:
             self.logger.error(f"Failed to calculate cost: {e}")
             raise ValueError(f"Failed to calculate cost: {e}") from e
@@ -158,7 +187,8 @@ class Client:
             "customerId": api_request.customer_id,
             "indicator": api_request.indicator,
             "amount": api_request.amount,
-            "inputToken": usage_data.prompt_tokens,
+            "inputToken": regular_prompt_tokens,  # Send non-cached tokens
+            "cachedToken": cached_tokens,  # Send cached tokens separately
             "outputToken": usage_data.completion_tokens,
             "model": usage_data.model,
             "serviceProvider": usage_data.service_provider

{paygent_sdk-2.0.0 → paygent_sdk-3.0.0}/paygent_sdk/constants.py

@@ -18,6 +18,7 @@ class ServiceProvider:
     MISTRAL_AI = "Mistral AI"
     COHERE = "Cohere"
     DEEPSEEK = "DeepSeek"
+    MOONSHOT_AI = "Moonshot AI"
     CUSTOM = "Custom"
     
     # STT Service Providers
@@ -25,12 +26,15 @@ class ServiceProvider:
     MICROSOFT_AZURE_SPEECH = "Microsoft Azure Speech Service"
     GOOGLE_CLOUD_SPEECH = "Google Cloud Speech-to-Text"
     ASSEMBLY_AI = "AssemblyAI"
+    ELEVEN_LABS_STT = "Eleven Labs STT"
+    SONIOX = "Soniox"
     
     # TTS Service Providers
     AMAZON_POLLY = "Amazon Polly"
     MICROSOFT_AZURE_SPEECH_TTS = "Microsoft Azure Speech Service"
     GOOGLE_CLOUD_TEXT_TO_SPEECH = "Google Cloud Text-to-Speech"
     DEEPGRAM_TTS = "Deepgram"
+    ELEVEN_LABS_TTS = "Eleven Labs TTS"
 
 
 # OpenAI Models
@@ -210,6 +214,13 @@ class DeepSeekModels:
     DEEPSEEK_V3_2_EXP = "DeepSeek V3.2-Exp"
 
 
+# Moonshot AI Models
+class MoonshotAIModels:
+    """Moonshot AI / Kimi model constants."""
+    KIMI_K2_INSTRUCT_0905 = "Kimi k2-instruct-0905"
+    KIMI_K2_0905_1T_256K = "Kimi k2-0905-1T-256K"
+
+
 
 
 
@@ -227,6 +238,11 @@ class DeepgramSTTModels:
     REDACTION = "Redaction (Add-on)"
     KEYTERM_PROMPTING = "Keyterm Prompting (Add-on)"
     SPEAKER_DIARIZATION = "Speaker Diarization (Add-on)"
+    # Growth tier models
+    GROWTH_NOVA_3_MONOLINGUAL = "Growth Nova-3 (Monolingual)"
+    GROWTH_NOVA_3_MULTILINGUAL = "Growth Nova-3 (Multilingual)"
+    GROWTH_NOVA_1 = "Growth Nova-1"
+    GROWTH_NOVA_2 = "Growth Nova-2"
 
 
 # Microsoft Azure Speech Service STT Models
@@ -250,6 +266,18 @@ class AssemblyAISTTModels:
     KEYTERMS_PROMPTING = "Keyterms Prompting"
 
 
+# Eleven Labs STT Models
+class ElevenLabsSTTModels:
+    """Eleven Labs STT model constants."""
+    BUSINESS_SCRIBE_V1_V2 = "Eleven Labs Business Scribe V1/V2"
+
+
+# Soniox STT Models
+class SonioxSTTModels:
+    """Soniox STT model constants."""
+    REAL_TIME = "Soniox Real Time"
+
+
 # Amazon Polly TTS Models
 class AmazonPollyTTSModels:
     """Amazon Polly TTS model constants."""
@@ -262,9 +290,9 @@ class AmazonPollyTTSModels:
 # Microsoft Azure Speech Service TTS Models
 class MicrosoftAzureSpeechTTSModels:
     """Microsoft Azure Speech Service TTS model constants."""
-    STANDARD = "Azure TTS Standard"
-    CUSTOM = "Azure TTS Custom"
-    CUSTOM_NEURAL_HD = "Azure TTS Custom Neural HD"
+    STANDARD_NEURAL = "Azure TTS Standard Neural"
+    CUSTOM_SYNTHESIS = "Azure TTS Custom Synthesis"
+    CUSTOM_SYNTHESIS_NEURAL_HD = "Azure TTS Custom Synthesis Neural HD"
 
 
 # Google Cloud Text-to-Speech TTS Models
@@ -284,6 +312,15 @@ class DeepgramTTSModels:
     """Deepgram TTS model constants."""
     AURA_2 = "Deepgram Aura-2"
     AURA_1 = "Deepgram Aura-1"
+    # Growth tier models
+    GROWTH_AURA_2 = "Deepgram Growth Aura-2"
+    GROWTH_AURA_1 = "Deepgram Growth Aura-1"
+
+
+# Eleven Labs TTS Models
+class ElevenLabsTTSModels:
+    """Eleven Labs TTS model constants."""
+    BUSINESS_MULTILINGUAL_V2_V3 = "Eleven Labs Business Multilingual V2/V3"
 
 
 def is_model_supported(model: str) -> bool:

{paygent_sdk-2.0.0 → paygent_sdk-3.0.0}/paygent_sdk/models.py

@@ -15,6 +15,7 @@ from .constants import (
     MistralAIModels,
     CohereModels,
     DeepSeekModels,
+    MoonshotAIModels,
 )
 
 
@@ -26,6 +27,7 @@ class UsageData:
     prompt_tokens: int
     completion_tokens: int
     total_tokens: int
+    cached_tokens: Optional[int] = None  # Optional cached tokens
 
 
 @dataclass
@@ -51,6 +53,7 @@ class ModelPricing:
     """Represents pricing information for different models."""
     prompt_tokens_cost: float
     completion_tokens_cost: float
+    cached_tokens_cost: Optional[float] = None  # Optional cached token cost (if None, model doesn't support caching)
 
 
 @dataclass
@@ -86,22 +89,27 @@ MODEL_PRICING: Dict[str, ModelPricing] = {
     # OpenAI Models (pricing per 1000 tokens)
     OpenAIModels.GPT_5: ModelPricing(
         prompt_tokens_cost=0.00125,  # $0.00125 per 1000 tokens
+        cached_tokens_cost=0.000125,  # 90% discount for cached tokens
         completion_tokens_cost=0.01  # $0.01 per 1000 tokens
     ),
     OpenAIModels.GPT_5_MINI: ModelPricing(
         prompt_tokens_cost=0.00025,  # $0.00025 per 1000 tokens
+        cached_tokens_cost=0.000025,  # 90% discount for cached tokens
         completion_tokens_cost=0.002  # $0.002 per 1000 tokens
     ),
     OpenAIModels.GPT_5_NANO: ModelPricing(
         prompt_tokens_cost=0.00005,  # $0.00005 per 1000 tokens
+        cached_tokens_cost=0.000005,  # 90% discount for cached tokens
         completion_tokens_cost=0.0004  # $0.0004 per 1000 tokens
     ),
     OpenAIModels.GPT_5_CHAT_LATEST: ModelPricing(
         prompt_tokens_cost=0.00125,  # $0.00125 per 1000 tokens
+        cached_tokens_cost=0.000125,  # 90% discount for cached tokens
         completion_tokens_cost=0.01  # $0.01 per 1000 tokens
     ),
     OpenAIModels.GPT_5_CODEX: ModelPricing(
         prompt_tokens_cost=0.00125,  # $0.00125 per 1000 tokens
+        cached_tokens_cost=0.000125,  # 90% discount for cached tokens
         completion_tokens_cost=0.01  # $0.01 per 1000 tokens
     ),
     OpenAIModels.GPT_5_PRO: ModelPricing(
@@ -114,26 +122,32 @@ MODEL_PRICING: Dict[str, ModelPricing] = {
     ),
     OpenAIModels.GPT_4_1: ModelPricing(
         prompt_tokens_cost=0.002,  # $0.002 per 1000 tokens
+        cached_tokens_cost=0.0005,  # 50% discount for cached tokens
         completion_tokens_cost=0.008  # $0.008 per 1000 tokens
     ),
     OpenAIModels.GPT_4_1_MINI: ModelPricing(
         prompt_tokens_cost=0.0004,  # $0.0004 per 1000 tokens
+        cached_tokens_cost=0.0001,  # 50% discount for cached tokens
         completion_tokens_cost=0.0016  # $0.0016 per 1000 tokens
     ),
     OpenAIModels.GPT_4_1_NANO: ModelPricing(
         prompt_tokens_cost=0.0001,  # $0.0001 per 1000 tokens
+        cached_tokens_cost=0.000025,  # 50% discount for cached tokens
         completion_tokens_cost=0.0004  # $0.0004 per 1000 tokens
     ),
     OpenAIModels.GPT_4O: ModelPricing(
         prompt_tokens_cost=0.0025,  # $0.0025 per 1000 tokens
+        cached_tokens_cost=0.00125,  # 50% discount for cached tok
         completion_tokens_cost=0.01  # $0.01 per 1000 tokens
     ),
     OpenAIModels.GPT_4O_2024_05_13: ModelPricing(
         prompt_tokens_cost=0.005,  # $0.005 per 1000 tokens
+        cached_tokens_cost=0.0025,  # 50% discount for cached tokens
         completion_tokens_cost=0.015  # $0.015 per 1000 tokens
     ),
     OpenAIModels.GPT_4O_MINI: ModelPricing(
         prompt_tokens_cost=0.00015,  # $0.00015 per 1000 tokens
+        cached_tokens_cost=0.000075,  # 50% discount for cached tokens
         completion_tokens_cost=0.0006  # $0.0006 per 1000 tokens
     ),
     OpenAIModels.GPT_REALTIME: ModelPricing(
@@ -146,10 +160,12 @@ MODEL_PRICING: Dict[str, ModelPricing] = {
     ),
     OpenAIModels.GPT_4O_REALTIME_PREVIEW: ModelPricing(
         prompt_tokens_cost=0.005,  # $0.005 per 1000 tokens
+        cached_tokens_cost=0.0025,  # 50% discount for cached tokens
         completion_tokens_cost=0.02  # $0.02 per 1000 tokens
     ),
     OpenAIModels.GPT_4O_MINI_REALTIME_PREVIEW: ModelPricing(
         prompt_tokens_cost=0.0006,  # $0.0006 per 1000 tokens
+        cached_tokens_cost=0.0003,  # 50% discount for cached tokens
         completion_tokens_cost=0.0024  # $0.0024 per 1000 tokens
     ),
     OpenAIModels.GPT_AUDIO: ModelPricing(
@@ -510,4 +526,16 @@ MODEL_PRICING: Dict[str, ModelPricing] = {
         prompt_tokens_cost=0.000028,  # $0.000028 per 1000 tokens
         completion_tokens_cost=0.00042  # $0.00042 per 1000 tokens
     ),
+
+    # Moonshot AI / Kimi Models (pricing per 1000 tokens)
+    MoonshotAIModels.KIMI_K2_INSTRUCT_0905: ModelPricing(
+        prompt_tokens_cost=0.001,  # $0.001 per 1000 tokens
+        cached_tokens_cost=0.0005,  # $0.0005 per 1000 tokens (as specified)
+        completion_tokens_cost=0.003  # $0.003 per 1000 tokens
+    ),
+    MoonshotAIModels.KIMI_K2_0905_1T_256K: ModelPricing(
+        prompt_tokens_cost=0.001,  # $0.001 per 1000 tokens
+        # cached_tokens_cost not specified - model doesn't support cached tokens
+        completion_tokens_cost=0.003  # $0.003 per 1000 tokens
+    ),
 }

{paygent_sdk-2.0.0 → paygent_sdk-3.0.0}/paygent_sdk/voice_client.py

@@ -13,10 +13,13 @@ from .constants import (
     MicrosoftAzureSpeechSTTModels,
     GoogleCloudSpeechSTTModels,
     AssemblyAISTTModels,
+    ElevenLabsSTTModels,
+    SonioxSTTModels,
     AmazonPollyTTSModels,
     MicrosoftAzureSpeechTTSModels,
     GoogleCloudTextToSpeechTTSModels,
     DeepgramTTSModels,
+    ElevenLabsTTSModels,
 )
 from .models import SttModelPricing, TtsModelPricing
 
@@ -34,6 +37,11 @@ STT_MODEL_PRICING: Dict[str, SttModelPricing] = {
     DeepgramSTTModels.REDACTION: SttModelPricing(cost_per_hour=0.12),  # $0.12 per hour (add-on)
     DeepgramSTTModels.KEYTERM_PROMPTING: SttModelPricing(cost_per_hour=0.072),  # $0.072 per hour (add-on)
     DeepgramSTTModels.SPEAKER_DIARIZATION: SttModelPricing(cost_per_hour=0.12),  # $0.12 per hour (add-on)
+    # Growth tier models
+    DeepgramSTTModels.GROWTH_NOVA_3_MONOLINGUAL: SttModelPricing(cost_per_hour=0.39),  # $0.39 per hour
+    DeepgramSTTModels.GROWTH_NOVA_3_MULTILINGUAL: SttModelPricing(cost_per_hour=0.468),  # $0.468 per hour
+    DeepgramSTTModels.GROWTH_NOVA_1: SttModelPricing(cost_per_hour=0.282),  # $0.282 per hour
+    DeepgramSTTModels.GROWTH_NOVA_2: SttModelPricing(cost_per_hour=0.282),  # $0.282 per hour
 
     # Microsoft Azure Speech Service Models
     MicrosoftAzureSpeechSTTModels.STANDARD: SttModelPricing(cost_per_hour=1.0),  # $1.0 per hour
@@ -46,6 +54,12 @@ STT_MODEL_PRICING: Dict[str, SttModelPricing] = {
     AssemblyAISTTModels.UNIVERSAL_STREAMING: SttModelPricing(cost_per_hour=0.15),  # $0.15 per hour
     AssemblyAISTTModels.UNIVERSAL_STREAMING_MULTILANG: SttModelPricing(cost_per_hour=0.15),  # $0.15 per hour
     AssemblyAISTTModels.KEYTERMS_PROMPTING: SttModelPricing(cost_per_hour=0.04),  # $0.04 per hour
+
+    # Eleven Labs STT Models
+    ElevenLabsSTTModels.BUSINESS_SCRIBE_V1_V2: SttModelPricing(cost_per_hour=0.22),  # $0.22 per hour
+
+    # Soniox STT Models
+    SonioxSTTModels.REAL_TIME: SttModelPricing(cost_per_hour=0.12),  # $0.12 per hour
 }
 
 
@@ -58,9 +72,9 @@ TTS_MODEL_PRICING: Dict[str, TtsModelPricing] = {
     AmazonPollyTTSModels.GENERATIVE: TtsModelPricing(cost_per_million_characters=30.0),  # $30 per 1 million characters
 
     # Microsoft Azure Speech Service TTS Models
-    MicrosoftAzureSpeechTTSModels.STANDARD: TtsModelPricing(cost_per_million_characters=15.0),  # $15 per 1 million characters
-    MicrosoftAzureSpeechTTSModels.CUSTOM: TtsModelPricing(cost_per_million_characters=24.0),  # $24 per 1 million characters
-    MicrosoftAzureSpeechTTSModels.CUSTOM_NEURAL_HD: TtsModelPricing(cost_per_million_characters=48.0),  # $48 per 1 million characters
+    MicrosoftAzureSpeechTTSModels.STANDARD_NEURAL: TtsModelPricing(cost_per_million_characters=15.0),  # $15 per 1 million characters
+    MicrosoftAzureSpeechTTSModels.CUSTOM_SYNTHESIS: TtsModelPricing(cost_per_million_characters=24.0),  # $24 per 1 million characters
+    MicrosoftAzureSpeechTTSModels.CUSTOM_SYNTHESIS_NEURAL_HD: TtsModelPricing(cost_per_million_characters=48.0),  # $48 per 1 million characters
 
     # Google Cloud Text-to-Speech TTS Models
     GoogleCloudTextToSpeechTTSModels.CHIRP_3_HD: TtsModelPricing(cost_per_million_characters=30.0),  # $30 per 1 million characters
@@ -74,6 +88,12 @@ TTS_MODEL_PRICING: Dict[str, TtsModelPricing] = {
     # Deepgram TTS Models
     DeepgramTTSModels.AURA_2: TtsModelPricing(cost_per_million_characters=30.0),  # $30 per 1 million characters
     DeepgramTTSModels.AURA_1: TtsModelPricing(cost_per_million_characters=15.0),  # $15 per 1 million characters
+    # Growth tier models
+    DeepgramTTSModels.GROWTH_AURA_2: TtsModelPricing(cost_per_million_characters=27.0),  # $27 per 1 million characters
+    DeepgramTTSModels.GROWTH_AURA_1: TtsModelPricing(cost_per_million_characters=13.5),  # $13.5 per 1 million characters
+
+    # Eleven Labs TTS Models
+    ElevenLabsTTSModels.BUSINESS_MULTILINGUAL_V2_V3: TtsModelPricing(cost_per_million_characters=120.0),  # $120 per 1 million characters
 }
 
 

paygent_sdk-3.0.0/paygent_sdk/wrappers/__init__.py

@@ -0,0 +1,44 @@
+"""
+Wrappers for automatic usage tracking with AI provider SDKs.
+
+This module provides wrapper classes that intercept API calls to various AI providers
+and automatically send usage data to Paygent for tracking and billing.
+
+Note: All wrappers are lazily imported to avoid requiring installation of peer dependencies
+that you don't use. For example, if you only use Gemini, you don't need to install openai,
+anthropic, or mistral packages.
+"""
+
+__all__ = [
+    "PaygentOpenAI",
+    "PaygentAnthropic",
+    "PaygentMistral",
+    "PaygentGemini",
+    "PaygentLangChainCallback",
+]
+
+
+def __getattr__(name):
+    """
+    Lazy import wrappers to avoid requiring peer dependencies that aren't being used.
+    
+    This allows users to only install the AI provider packages they actually use,
+    rather than requiring all of them as dependencies.
+    """
+    if name == "PaygentOpenAI":
+        from .openai_wrapper import PaygentOpenAI
+        return PaygentOpenAI
+    elif name == "PaygentAnthropic":
+        from .anthropic_wrapper import PaygentAnthropic
+        return PaygentAnthropic
+    elif name == "PaygentMistral":
+        from .mistral_wrapper import PaygentMistral
+        return PaygentMistral
+    elif name == "PaygentGemini":
+        from .gemini_wrapper import PaygentGemini
+        return PaygentGemini
+    elif name == "PaygentLangChainCallback":
+        from .langchain_wrapper import PaygentLangChainCallback
+        return PaygentLangChainCallback
+    
+    raise AttributeError(f"module '{__name__}' has no attribute '{name}'")

{paygent_sdk-2.0.0 → paygent_sdk-3.0.0}/paygent_sdk/wrappers/openai_wrapper.py

@@ -113,12 +113,20 @@ class ChatCompletionsWrapper:
 
         if has_valid_usage:
             # Primary path: Use usage data from API response
+            # Extract cached tokens if available (OpenAI prompt caching feature)
+            cached_tokens = 0
+            if hasattr(response.usage, 'prompt_tokens_details'):
+                prompt_details = response.usage.prompt_tokens_details
+                if hasattr(prompt_details, 'cached_tokens') and prompt_details.cached_tokens:
+                    cached_tokens = prompt_details.cached_tokens
+            
             usage_data = UsageData(
                 service_provider=model,
                 model=model,
                 prompt_tokens=response.usage.prompt_tokens,
                 completion_tokens=response.usage.completion_tokens,
-                total_tokens=response.usage.total_tokens
+                total_tokens=response.usage.total_tokens,
+                cached_tokens=cached_tokens
             )
 
             self.paygent_client.send_usage(

{paygent_sdk-2.0.0 → paygent_sdk-3.0.0/paygent_sdk.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: paygent-sdk
-Version: 2.0.0
+Version: 3.0.0
 Summary: Official Python SDK for Paygent - Track AI usage and costs across multiple providers (OpenAI, Anthropic, Google, DeepSeek, etc.)
 Home-page: https://github.com/paygent/paygent-sdk-python
 Author: Paygent

{paygent_sdk-2.0.0 → paygent_sdk-3.0.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "paygent-sdk"
-version = "2.0.0"
+version = "3.0.0"
 description = "Official Python SDK for Paygent - Track AI usage and costs across multiple providers (OpenAI, Anthropic, Google, DeepSeek, etc.)"
 readme = "README.md"
 requires-python = ">=3.7"

paygent_sdk-2.0.0/paygent_sdk/wrappers/__init__.py

@@ -1,30 +0,0 @@
-"""
-Wrappers for automatic usage tracking with AI provider SDKs.
-
-This module provides wrapper classes that intercept API calls to various AI providers
-and automatically send usage data to Paygent for tracking and billing.
-"""
-
-from .openai_wrapper import PaygentOpenAI
-from .anthropic_wrapper import PaygentAnthropic
-from .mistral_wrapper import PaygentMistral
-from .gemini_wrapper import PaygentGemini
-
-# LangChain integration (optional dependency)
-try:
-    from .langchain_wrapper import PaygentLangChainCallback
-    __all__ = [
-        "PaygentOpenAI",
-        "PaygentAnthropic",
-        "PaygentMistral",
-        "PaygentGemini",
-        "PaygentLangChainCallback",
-    ]
-except ImportError:
-    # LangChain not installed
-    __all__ = [
-        "PaygentOpenAI",
-        "PaygentAnthropic",
-        "PaygentMistral",
-        "PaygentGemini",
-    ]