ai-pipeline-core 0.2.1__tar.gz → 0.2.2__tar.gz

{ai_pipeline_core-0.2.1 → ai_pipeline_core-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ai-pipeline-core
-Version: 0.2.1
+Version: 0.2.2
 Summary: Core utilities for AI-powered processing pipelines using prefect
 Project-URL: Homepage, https://github.com/bbarwik/ai-pipeline-core
 Project-URL: Repository, https://github.com/bbarwik/ai-pipeline-core

{ai_pipeline_core-0.2.1 → ai_pipeline_core-0.2.2}/ai_pipeline_core/__init__.py

@@ -118,7 +118,7 @@ from .prompt_manager import PromptManager
 from .settings import Settings
 from .tracing import TraceInfo, TraceLevel, set_trace_cost, trace
 
-__version__ = "0.2.1"
+__version__ = "0.2.2"
 
 __all__ = [
     # Config/Settings

{ai_pipeline_core-0.2.1 → ai_pipeline_core-0.2.2}/ai_pipeline_core/llm/client.py

@@ -37,7 +37,7 @@ def _process_messages(
     context: AIMessages,
     messages: AIMessages,
     system_prompt: str | None = None,
-    cache_ttl: str | None = "120s",
+    cache_ttl: str | None = "5m",
 ) -> list[ChatCompletionMessageParam]:
     """Process and format messages for LLM API consumption.
 
@@ -245,7 +245,7 @@ async def generate(
         model: Model to use (e.g., "gpt-5", "gemini-2.5-pro", "grok-4").
                Accepts predefined models or any string for custom models.
         context: Static context to cache (documents, examples, instructions).
-                Defaults to None (empty context). Cached for 120 seconds.
+                Defaults to None (empty context). Cached for 5 minutes by default.
         messages: Dynamic messages/queries. AIMessages or str ONLY.
                  Do not pass Document or DocumentList directly.
                  If string, converted to AIMessages internally.
@@ -338,13 +338,13 @@ async def generate(
         - Context caching saves ~50-90% tokens on repeated calls
         - First call: full token cost
         - Subsequent calls (within cache TTL): only messages tokens
-        - Default cache TTL is 120s (production-optimized)
+        - Default cache TTL is 5m (production-optimized)
         - Default retry logic: 3 attempts with 10s delay (production-optimized)
 
     Caching:
         When enabled in your LiteLLM proxy and supported by the upstream provider,
         context messages may be cached to reduce token usage on repeated calls.
-        Default TTL is 120s (optimized for production workloads). Configure caching
+        Default TTL is 5m (optimized for production workloads). Configure caching
         behavior centrally via your LiteLLM proxy settings, not per API call.
         Savings depend on provider and payload; treat this as an optimization, not a guarantee.
 

{ai_pipeline_core-0.2.1 → ai_pipeline_core-0.2.2}/ai_pipeline_core/llm/model_options.py

@@ -45,7 +45,7 @@ class ModelOptions(BaseModel):
 
         timeout: Maximum seconds to wait for response (default: 300).
 
-        cache_ttl: Cache TTL for context messages (default: "120s").
+        cache_ttl: Cache TTL for context messages (default: "5m").
                    String format like "60s", "5m", or None to disable caching.
                    Applied to the last context message for efficient token reuse.
 
@@ -109,7 +109,7 @@ class ModelOptions(BaseModel):
         - search_context_size only works with search models
         - reasoning_effort only works with models that support explicit reasoning
         - response_format is set internally by generate_structured()
-        - cache_ttl accepts formats like "120s", "5m", "1h" or None to disable caching
+        - cache_ttl accepts formats like "120s", "5m" (default), "1h" or None to disable caching
     """
 
     temperature: float | None = None
@@ -118,11 +118,13 @@ class ModelOptions(BaseModel):
     reasoning_effort: Literal["low", "medium", "high"] | None = None
     retries: int = 3
     retry_delay_seconds: int = 10
-    timeout: int = 300
-    cache_ttl: str | None = "120s"
+    timeout: int = 600
+    cache_ttl: str | None = "5m"
     service_tier: Literal["auto", "default", "flex", "scale", "priority"] | None = None
     max_completion_tokens: int | None = None
     response_format: type[BaseModel] | None = None
+    verbosity: Literal["low", "medium", "high"] | None = None
+    usage_tracking: bool = True
 
     def to_openai_completion_kwargs(self) -> dict[str, Any]:
         """Convert options to OpenAI API completion parameters.
@@ -163,9 +165,7 @@ class ModelOptions(BaseModel):
         """
         kwargs: dict[str, Any] = {
             "timeout": self.timeout,
-            "extra_body": {
-                "usage": {"include": True},  # For openrouter cost tracking
-            },
+            "extra_body": {},
         }
 
         if self.temperature:
@@ -186,4 +186,10 @@ class ModelOptions(BaseModel):
         if self.service_tier:
             kwargs["service_tier"] = self.service_tier
 
+        if self.verbosity:
+            kwargs["verbosity"] = self.verbosity
+
+        if self.usage_tracking:
+            kwargs["extra_body"]["usage"] = {"include": True}
+
         return kwargs

{ai_pipeline_core-0.2.1 → ai_pipeline_core-0.2.2}/ai_pipeline_core/llm/model_response.py

@@ -110,7 +110,8 @@ class ModelResponse(ChatCompletion):
             >>> if "error" in response.content.lower():
             ...     # Handle error case
         """
-        return self.choices[0].message.content or ""
+        content = self.choices[0].message.content or ""
+        return content.split("</think>")[-1].strip()
 
     def set_model_options(self, options: dict[str, Any]) -> None:
         """Store the model configuration used for generation.

{ai_pipeline_core-0.2.1 → ai_pipeline_core-0.2.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ai-pipeline-core"
-version = "0.2.1"
+version = "0.2.2"
 description = "Core utilities for AI-powered processing pipelines using prefect"
 readme = "README.md"
 license = {text = "MIT"}
@@ -177,7 +177,7 @@ reportIncompatibleVariableOverride = "error"
 reportMissingParameterType = "warning"
 
 [tool.bumpversion]
-current_version = "0.2.1"
+current_version = "0.2.2"
 commit = true
 tag = true
 tag_name = "v{new_version}"