keycycle 0.2.2__tar.gz → 0.2.4__tar.gz

{keycycle-0.2.2 → keycycle-0.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keycycle
-Version: 0.2.2
+Version: 0.2.4
 Summary: A thread-safe key rotation and rate limiting manager for API keys.
 Author-email: William Jin <jinwilliam.jin@gmail.com>
 Classifier: Programming Language :: Python :: 3

{keycycle-0.2.2 → keycycle-0.2.4}/keycycle/adapters/openai_adapter.py

@@ -37,7 +37,7 @@ PROVIDER_BASE_URLS = {
     "cerebras": "https://api.cerebras.ai/v1",
     "groq": "https://api.groq.com/openai/v1",
     "cohere": "https://api.cohere.ai/compatibility/v1",
-
+    "moonshot": "https://api.moonshot.ai/v1",
 }
 
 class BaseRotatingClient:

keycycle-0.2.4/keycycle/config/models/moonshot.yaml

@@ -0,0 +1,59 @@
+# Moonshot API rate limits (dummy high limits)
+# Base URL: https://api.moonshot.ai/v1
+# Agno class: MoonShot (similar to OpenRouter special case)
+
+default:
+  requests_per_minute: 10000
+  requests_per_hour: 100000
+  requests_per_day: 1000000
+  tokens_per_minute: 10000000
+  tokens_per_hour: 100000000
+  tokens_per_day: 1000000000
+
+kimi-k2.5:
+  requests_per_minute: 10000
+  requests_per_hour: 100000
+  requests_per_day: 1000000
+  tokens_per_minute: 10000000
+  tokens_per_hour: 100000000
+  tokens_per_day: 1000000000
+
+kimi-k2-0905-preview:
+  requests_per_minute: 10000
+  requests_per_hour: 100000
+  requests_per_day: 1000000
+  tokens_per_minute: 10000000
+  tokens_per_hour: 100000000
+  tokens_per_day: 1000000000
+
+kimi-k2-0711-preview:
+  requests_per_minute: 10000
+  requests_per_hour: 100000
+  requests_per_day: 1000000
+  tokens_per_minute: 10000000
+  tokens_per_hour: 100000000
+  tokens_per_day: 1000000000
+
+kimi-k2-turbo-preview:
+  requests_per_minute: 10000
+  requests_per_hour: 100000
+  requests_per_day: 1000000
+  tokens_per_minute: 10000000
+  tokens_per_hour: 100000000
+  tokens_per_day: 1000000000
+
+kimi-k2-thinking:
+  requests_per_minute: 10000
+  requests_per_hour: 100000
+  requests_per_day: 1000000
+  tokens_per_minute: 10000000
+  tokens_per_hour: 100000000
+  tokens_per_day: 1000000000
+
+kimi-k2-thinking-turbo:
+  requests_per_minute: 10000
+  requests_per_hour: 100000
+  requests_per_day: 1000000
+  tokens_per_minute: 10000000
+  tokens_per_hour: 100000000
+  tokens_per_day: 1000000000

{keycycle-0.2.2 → keycycle-0.2.4}/keycycle/config/models.py

@@ -15,13 +15,15 @@ class ModelDict(TypedDict):
     gemini: Any
     openrouter: Any
     cohere: Any
+    moonshot: Any
 
 PROVIDER_STRATEGIES: ModelDict = {
     'cerebras': RateLimitStrategy.PER_MODEL,
     'groq': RateLimitStrategy.PER_MODEL,
     'gemini': RateLimitStrategy.PER_MODEL,
     'openrouter': RateLimitStrategy.GLOBAL,
-    'cohere': RateLimitStrategy.PER_MODEL
+    'cohere': RateLimitStrategy.PER_MODEL,
+    'moonshot': RateLimitStrategy.PER_MODEL
 }
 
 # Load Cohere tiers first to handle the env var logic
@@ -41,7 +43,8 @@ MODEL_LIMITS: ModelDict = {
     'cohere':{
         # Same for every model
         'default': COHERE_TIERS.get(os.getenv('COHERE_TIER', 'free').lower(), COHERE_TIERS['free'])
-    }
+    },
+    'moonshot': load_rate_limits_from_yaml(os.path.join(MODELS_DIR, 'moonshot.yaml'))
 }
 
 OPENROUTER_MODELS = load_openrouter_models(os.path.join(MODELS_DIR, 'openrouter_models.yaml'))

{keycycle-0.2.2 → keycycle-0.2.4}/keycycle/key_rotation/rotating_mixin.py

@@ -14,7 +14,17 @@ from ..config.constants import (
 )
 from ..core.utils import is_rate_limit_error, is_temporary_rate_limit_error, get_key_suffix
 from ..core.backoff import ExponentialBackoff, BackoffConfig
-from agno.models.response import ModelResponse
+if TYPE_CHECKING:
+    from agno.models.response import ModelResponse
+
+
+class _UsageResponse:
+    """Minimal response holder for usage recording when agno is not available."""
+    __slots__ = ('response_usage',)
+
+    def __init__(self, usage=None):
+        self.response_usage = usage
+
 
 class RotatingCredentialsMixin:
     """
@@ -100,7 +110,7 @@ class RotatingCredentialsMixin:
     def _get_retry_limit(self) -> int:
         return min(self._max_retries, len(self.wrapper.manager.keys) - 1)
 
-    def _record_usage(self, key_obj: KeyUsage, response: Optional[ModelResponse]):
+    def _record_usage(self, key_obj: KeyUsage, response: Optional["ModelResponse"]):
         """
         Extracts usage from the response and reports it to the manager.
         Falls back to estimated_tokens if response is None or usage is missing.
@@ -132,7 +142,7 @@ class RotatingCredentialsMixin:
             multiplier=TEMP_RATE_LIMIT_MULTIPLIER,
         ))
 
-    def invoke(self, *args, **kwargs) -> ModelResponse:
+    def invoke(self, *args, **kwargs) -> "ModelResponse":
         limit = self._get_retry_limit()
 
         for attempt in range(limit + 1):
@@ -178,7 +188,7 @@ class RotatingCredentialsMixin:
                     continue
                 raise
 
-    async def ainvoke(self, *args, **kwargs) -> ModelResponse:
+    async def ainvoke(self, *args, **kwargs) -> "ModelResponse":
         limit = self._get_retry_limit()
 
         for attempt in range(limit + 1):
@@ -224,7 +234,7 @@ class RotatingCredentialsMixin:
                     continue
                 raise
     
-    def invoke_stream(self, *args, **kwargs) -> Iterator[ModelResponse]:
+    def invoke_stream(self, *args, **kwargs) -> Iterator["ModelResponse"]:
         limit = self._get_retry_limit()
 
         for attempt in range(limit + 1):
@@ -242,8 +252,7 @@ class RotatingCredentialsMixin:
                             final_usage = chunk.response_usage
                         yield chunk
                     if final_usage:
-                        dummy_response = ModelResponse()
-                        dummy_response.response_usage = final_usage
+                        dummy_response = _UsageResponse(final_usage)
                         self._record_usage(key_usage, dummy_response)
                     else:
                         # If the stream didn't return usage data, fallback to estimation
@@ -280,7 +289,7 @@ class RotatingCredentialsMixin:
                     continue
                 raise
 
-    async def ainvoke_stream(self, *args, **kwargs) -> AsyncIterator[ModelResponse]:
+    async def ainvoke_stream(self, *args, **kwargs) -> AsyncIterator["ModelResponse"]:
         limit = self._get_retry_limit()
 
         for attempt in range(limit + 1):
@@ -301,8 +310,7 @@ class RotatingCredentialsMixin:
 
                     # Stream completed successfully. Record usage.
                     if final_usage:
-                        dummy_response = ModelResponse()
-                        dummy_response.response_usage = final_usage
+                        dummy_response = _UsageResponse(final_usage)
                         self._record_usage(key_usage, dummy_response)
                     else:
                         self._record_usage(key_usage, None)

{keycycle-0.2.2 → keycycle-0.2.4}/keycycle/utils.py

@@ -12,7 +12,8 @@ def get_agno_model_class(provider: str):
         "gemini": "Gemini",
         "azure": "AzureOpenAI",
         "aws": "Bedrock",
-        "openrouter": "OpenRouter"
+        "openrouter": "OpenRouter",
+        "moonshot": "MoonShot"
     }
 
     class_name = overrides.get(p_low, p_low.capitalize())

{keycycle-0.2.2 → keycycle-0.2.4}/keycycle.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keycycle
-Version: 0.2.2
+Version: 0.2.4
 Summary: A thread-safe key rotation and rate limiting manager for API keys.
 Author-email: William Jin <jinwilliam.jin@gmail.com>
 Classifier: Programming Language :: Python :: 3

{keycycle-0.2.2 → keycycle-0.2.4}/keycycle.egg-info/SOURCES.txt

@@ -24,6 +24,7 @@ keycycle/config/models/cerebras.yaml
 keycycle/config/models/cohere.yaml
 keycycle/config/models/gemini.yaml
 keycycle/config/models/groq.yaml
+keycycle/config/models/moonshot.yaml
 keycycle/config/models/openrouter.yaml
 keycycle/config/models/openrouter_models.yaml
 keycycle/core/__init__.py

{keycycle-0.2.2 → keycycle-0.2.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "keycycle"
-version = "0.2.2"
+version = "0.2.4"
 authors = [
   { name="William Jin", email="jinwilliam.jin@gmail.com" },
 ]
@@ -117,5 +117,6 @@ module = [
     "groq.*",
     "cerebras.*",
     "openrouter.*",
+    "moonshot.*",
 ]
 ignore_missing_imports = true