ai-pipeline-core 0.1.6__tar.gz → 0.1.8__tar.gz

{ai_pipeline_core-0.1.6 → ai_pipeline_core-0.1.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ai-pipeline-core
-Version: 0.1.6
+Version: 0.1.8
 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
@@ -20,7 +20,7 @@ Classifier: Typing :: Typed
 Requires-Python: >=3.12
 Requires-Dist: httpx>=0.28.1
 Requires-Dist: jinja2>=3.1.6
-Requires-Dist: lmnr>=0.7.5
+Requires-Dist: lmnr>=0.7.6
 Requires-Dist: openai>=1.99.9
 Requires-Dist: prefect>=3.4.13
 Requires-Dist: pydantic-settings>=2.10.1
@@ -151,40 +151,76 @@ async def process_document(doc: Document):
     return response.parsed
 ```
 
-### Prefect Flow Integration
+### Enhanced Pipeline Decorators
 ```python
-from prefect import flow, task
-from ai_pipeline_core.documents import Document, DocumentList, FlowDocument
-from ai_pipeline_core.flow import FlowConfig
-from ai_pipeline_core.tracing import trace
-
-class OutputDocument(FlowDocument):
-    """Custom output document type"""
-    def get_type(self) -> str:
-        return "output"
+from ai_pipeline_core import pipeline_flow, pipeline_task
+from ai_pipeline_core.flow import FlowOptions
+from ai_pipeline_core.documents import DocumentList, FlowDocument
 
-class MyFlowConfig(FlowConfig):
-    INPUT_DOCUMENT_TYPES = [InputDocument]
-    OUTPUT_DOCUMENT_TYPE = OutputDocument
+class CustomFlowOptions(FlowOptions):
+    """Extend base options with your custom fields"""
+    batch_size: int = 100
+    temperature: float = 0.7
 
-@task
-@trace
+@pipeline_task(trace_level="always", retries=3)
 async def process_task(doc: Document) -> Document:
-    # Task-level processing with automatic tracing
+    # Task with automatic tracing and retries
     result = await process_document(doc)
-    # Convert result to JSON string for document content
-    import json
-    return OutputDocument(name="result", content=json.dumps(result.model_dump()).encode())
+    return OutputDocument(name="result", content=result.encode())
+
+@pipeline_flow(trace_level="always")
+async def my_pipeline(
+    project_name: str,
+    documents: DocumentList,
+    flow_options: CustomFlowOptions  # Type-safe custom options
+) -> DocumentList:
+    # Pipeline flow with enforced signature and tracing
+    results = []
+    for doc in documents:
+        result = await process_task(doc)
+        results.append(result)
+    return DocumentList(results)
+```
 
-@flow
-async def my_pipeline(documents: DocumentList):
-    config = MyFlowConfig()
-    input_docs = config.get_input_documents(documents)
+### Simple Runner Utility
+```python
+from ai_pipeline_core.simple_runner import run_cli, run_pipeline
+from ai_pipeline_core.flow import FlowOptions
+
+# CLI-based pipeline execution
+if __name__ == "__main__":
+    run_cli(
+        flows=[my_pipeline],
+        flow_configs=[MyFlowConfig],
+        options_cls=CustomFlowOptions
+    )
 
-    results = await process_task.map(input_docs)
+# Or programmatic execution
+async def main():
+    result = await run_pipeline(
+        project_name="my-project",
+        output_dir=Path("./output"),
+        flow=my_pipeline,
+        flow_config=MyFlowConfig,
+        flow_options=CustomFlowOptions(batch_size=50)
+    )
+```
+
+### Clean Prefect Decorators
+```python
+# Import clean Prefect decorators without tracing
+from ai_pipeline_core.prefect import flow, task
 
-    config.validate_output_documents(results)
-    return results
+# Or use pipeline decorators with tracing
+from ai_pipeline_core import pipeline_flow, pipeline_task
+
+@task  # Clean Prefect task (supports both sync and async)
+def compute(x: int) -> int:
+    return x * 2
+
+@pipeline_task(trace_level="always")  # With tracing (async only)
+async def compute_traced(x: int) -> int:
+    return x * 2
 ```
 
 ## Core Modules
@@ -210,12 +246,12 @@ docs = DocumentList([doc1, doc2])
 Managed AI interactions with built-in retry logic, cost tracking, and structured outputs.
 
 **Supported Models** (via LiteLLM proxy):
-- OpenAI: GPT-4, GPT-5 series
-- Anthropic: Claude 3 series
-- Google: Gemini 2.5 series
-- xAI: Grok models
-- Perplexity: Sonar models (with search capabilities)
-- And many more through LiteLLM compatibility
+- OpenAI: gpt-5
+- Anthropic: claude-4
+- Google: gemini-2.5
+- xAI: grok-3, grok-4
+- Perplexity: sonar-pro-search
+- And many more through LiteLLM compatibility. Every model from openrouter should work.
 
 ```python
 from ai_pipeline_core.llm import generate_structured, AIMessages, ModelOptions
@@ -291,8 +327,14 @@ ai_pipeline_core/
 │   ├── client.py      # Async client implementation
 │   └── model_options.py # Configuration models
 ├── flow/              # Prefect flow utilities
-│   └── config.py      # Type-safe flow configuration
+│   ├── config.py      # Type-safe flow configuration
+│   └── options.py     # FlowOptions base class
+├── simple_runner/     # Pipeline execution utilities
+│   ├── cli.py         # CLI interface
+│   └── simple_runner.py # Core runner logic
 ├── logging/           # Structured logging
+├── pipeline.py        # Enhanced decorators
+├── prefect.py         # Clean Prefect exports
 ├── tracing.py         # Observability decorators
 └── settings.py        # Centralized configuration
 ```
@@ -303,6 +345,7 @@ ai_pipeline_core/
 ```bash
 make test           # Run all tests
 make test-cov      # Run with coverage report
+make test-showcase # Test the showcase.py CLI example
 pytest tests/test_documents.py::TestDocument::test_creation  # Single test
 ```
 
@@ -439,6 +482,22 @@ For learning purposes, see [CLAUDE.md](CLAUDE.md) for our comprehensive coding s
 
 - [CLAUDE.md](CLAUDE.md) - Detailed coding standards and architecture guide
 
+## Examples
+
+### In This Repository
+- [showcase.py](examples/showcase.py) - Complete example demonstrating all core features including the CLI runner
+  ```bash
+  # Run the showcase example with CLI
+  python examples/showcase.py ./output --temperature 0.7 --batch-size 5
+
+  # Show help
+  python examples/showcase.py --help
+  ```
+- [showcase.jinja2](examples/showcase.jinja2) - Example Jinja2 prompt template
+
+### Real-World Application
+- [AI Documentation Writer](https://github.com/bbarwik/ai-documentation-writer) - Production-ready example showing how to build sophisticated AI pipelines for automated documentation generation. See [examples/ai-documentation-writer.md](examples/ai-documentation-writer.md) for a detailed overview.
+
 ### dependencies_docs/ Directory
 > [!NOTE]
 > The `dependencies_docs/` directory contains guides for AI assistants (like Claude Code) on how to interact with the project's external dependencies and tooling, NOT user documentation for ai-pipeline-core itself. These files are excluded from repository listings to avoid confusion.
@@ -469,9 +528,29 @@ Built with:
 - [LiteLLM](https://litellm.ai/) - LLM proxy
 - [Pydantic](https://pydantic-docs.helpmanual.io/) - Data validation
 
+## What's New in v0.1.8
+
+### Breaking Changes
+- **Async-Only Pipeline Decorators**: `@pipeline_flow` and `@pipeline_task` now require `async def` functions (raises TypeError for sync)
+- **Document Class Name Validation**: Document subclasses cannot start with "Test" prefix (pytest conflict prevention)
+- **FlowConfig Validation**: OUTPUT_DOCUMENT_TYPE cannot be in INPUT_DOCUMENT_TYPES (prevents circular dependencies)
+- **Temperature Field**: Added optional `temperature` field to `ModelOptions` for explicit control
+
+### Major Improvements
+- **Pipeline Module Refactoring**: Reduced from ~400 to ~150 lines with cleaner Protocol-based typing
+- **Enhanced Validation**: FlowConfig and Document classes now validate at definition time
+- **Better CLI Support**: Auto-displays help when no arguments provided, improved context management
+- **Test Suite Updates**: All tests updated to use async/await consistently
+
+### Documentation Updates
+- Added Document naming rules to CLAUDE.md
+- Added FlowConfig validation rules
+- Added code elegance principles section
+- Updated guide_for_ai.md to API reference format
+
 ## Stability Notice
 
-**Current Version**: 0.1.6
+**Current Version**: 0.1.8
 **Status**: Internal Preview
 **API Stability**: Unstable - Breaking changes expected
 **Recommended Use**: Learning and reference only

{ai_pipeline_core-0.1.6 → ai_pipeline_core-0.1.8}/README.md

@@ -109,40 +109,76 @@ async def process_document(doc: Document):
     return response.parsed
 ```
 
-### Prefect Flow Integration
+### Enhanced Pipeline Decorators
 ```python
-from prefect import flow, task
-from ai_pipeline_core.documents import Document, DocumentList, FlowDocument
-from ai_pipeline_core.flow import FlowConfig
-from ai_pipeline_core.tracing import trace
-
-class OutputDocument(FlowDocument):
-    """Custom output document type"""
-    def get_type(self) -> str:
-        return "output"
+from ai_pipeline_core import pipeline_flow, pipeline_task
+from ai_pipeline_core.flow import FlowOptions
+from ai_pipeline_core.documents import DocumentList, FlowDocument
 
-class MyFlowConfig(FlowConfig):
-    INPUT_DOCUMENT_TYPES = [InputDocument]
-    OUTPUT_DOCUMENT_TYPE = OutputDocument
+class CustomFlowOptions(FlowOptions):
+    """Extend base options with your custom fields"""
+    batch_size: int = 100
+    temperature: float = 0.7
 
-@task
-@trace
+@pipeline_task(trace_level="always", retries=3)
 async def process_task(doc: Document) -> Document:
-    # Task-level processing with automatic tracing
+    # Task with automatic tracing and retries
     result = await process_document(doc)
-    # Convert result to JSON string for document content
-    import json
-    return OutputDocument(name="result", content=json.dumps(result.model_dump()).encode())
+    return OutputDocument(name="result", content=result.encode())
+
+@pipeline_flow(trace_level="always")
+async def my_pipeline(
+    project_name: str,
+    documents: DocumentList,
+    flow_options: CustomFlowOptions  # Type-safe custom options
+) -> DocumentList:
+    # Pipeline flow with enforced signature and tracing
+    results = []
+    for doc in documents:
+        result = await process_task(doc)
+        results.append(result)
+    return DocumentList(results)
+```
 
-@flow
-async def my_pipeline(documents: DocumentList):
-    config = MyFlowConfig()
-    input_docs = config.get_input_documents(documents)
+### Simple Runner Utility
+```python
+from ai_pipeline_core.simple_runner import run_cli, run_pipeline
+from ai_pipeline_core.flow import FlowOptions
+
+# CLI-based pipeline execution
+if __name__ == "__main__":
+    run_cli(
+        flows=[my_pipeline],
+        flow_configs=[MyFlowConfig],
+        options_cls=CustomFlowOptions
+    )
 
-    results = await process_task.map(input_docs)
+# Or programmatic execution
+async def main():
+    result = await run_pipeline(
+        project_name="my-project",
+        output_dir=Path("./output"),
+        flow=my_pipeline,
+        flow_config=MyFlowConfig,
+        flow_options=CustomFlowOptions(batch_size=50)
+    )
+```
+
+### Clean Prefect Decorators
+```python
+# Import clean Prefect decorators without tracing
+from ai_pipeline_core.prefect import flow, task
 
-    config.validate_output_documents(results)
-    return results
+# Or use pipeline decorators with tracing
+from ai_pipeline_core import pipeline_flow, pipeline_task
+
+@task  # Clean Prefect task (supports both sync and async)
+def compute(x: int) -> int:
+    return x * 2
+
+@pipeline_task(trace_level="always")  # With tracing (async only)
+async def compute_traced(x: int) -> int:
+    return x * 2
 ```
 
 ## Core Modules
@@ -168,12 +204,12 @@ docs = DocumentList([doc1, doc2])
 Managed AI interactions with built-in retry logic, cost tracking, and structured outputs.
 
 **Supported Models** (via LiteLLM proxy):
-- OpenAI: GPT-4, GPT-5 series
-- Anthropic: Claude 3 series
-- Google: Gemini 2.5 series
-- xAI: Grok models
-- Perplexity: Sonar models (with search capabilities)
-- And many more through LiteLLM compatibility
+- OpenAI: gpt-5
+- Anthropic: claude-4
+- Google: gemini-2.5
+- xAI: grok-3, grok-4
+- Perplexity: sonar-pro-search
+- And many more through LiteLLM compatibility. Every model from openrouter should work.
 
 ```python
 from ai_pipeline_core.llm import generate_structured, AIMessages, ModelOptions
@@ -249,8 +285,14 @@ ai_pipeline_core/
 │   ├── client.py      # Async client implementation
 │   └── model_options.py # Configuration models
 ├── flow/              # Prefect flow utilities
-│   └── config.py      # Type-safe flow configuration
+│   ├── config.py      # Type-safe flow configuration
+│   └── options.py     # FlowOptions base class
+├── simple_runner/     # Pipeline execution utilities
+│   ├── cli.py         # CLI interface
+│   └── simple_runner.py # Core runner logic
 ├── logging/           # Structured logging
+├── pipeline.py        # Enhanced decorators
+├── prefect.py         # Clean Prefect exports
 ├── tracing.py         # Observability decorators
 └── settings.py        # Centralized configuration
 ```
@@ -261,6 +303,7 @@ ai_pipeline_core/
 ```bash
 make test           # Run all tests
 make test-cov      # Run with coverage report
+make test-showcase # Test the showcase.py CLI example
 pytest tests/test_documents.py::TestDocument::test_creation  # Single test
 ```
 
@@ -397,6 +440,22 @@ For learning purposes, see [CLAUDE.md](CLAUDE.md) for our comprehensive coding s
 
 - [CLAUDE.md](CLAUDE.md) - Detailed coding standards and architecture guide
 
+## Examples
+
+### In This Repository
+- [showcase.py](examples/showcase.py) - Complete example demonstrating all core features including the CLI runner
+  ```bash
+  # Run the showcase example with CLI
+  python examples/showcase.py ./output --temperature 0.7 --batch-size 5
+
+  # Show help
+  python examples/showcase.py --help
+  ```
+- [showcase.jinja2](examples/showcase.jinja2) - Example Jinja2 prompt template
+
+### Real-World Application
+- [AI Documentation Writer](https://github.com/bbarwik/ai-documentation-writer) - Production-ready example showing how to build sophisticated AI pipelines for automated documentation generation. See [examples/ai-documentation-writer.md](examples/ai-documentation-writer.md) for a detailed overview.
+
 ### dependencies_docs/ Directory
 > [!NOTE]
 > The `dependencies_docs/` directory contains guides for AI assistants (like Claude Code) on how to interact with the project's external dependencies and tooling, NOT user documentation for ai-pipeline-core itself. These files are excluded from repository listings to avoid confusion.
@@ -427,9 +486,29 @@ Built with:
 - [LiteLLM](https://litellm.ai/) - LLM proxy
 - [Pydantic](https://pydantic-docs.helpmanual.io/) - Data validation
 
+## What's New in v0.1.8
+
+### Breaking Changes
+- **Async-Only Pipeline Decorators**: `@pipeline_flow` and `@pipeline_task` now require `async def` functions (raises TypeError for sync)
+- **Document Class Name Validation**: Document subclasses cannot start with "Test" prefix (pytest conflict prevention)
+- **FlowConfig Validation**: OUTPUT_DOCUMENT_TYPE cannot be in INPUT_DOCUMENT_TYPES (prevents circular dependencies)
+- **Temperature Field**: Added optional `temperature` field to `ModelOptions` for explicit control
+
+### Major Improvements
+- **Pipeline Module Refactoring**: Reduced from ~400 to ~150 lines with cleaner Protocol-based typing
+- **Enhanced Validation**: FlowConfig and Document classes now validate at definition time
+- **Better CLI Support**: Auto-displays help when no arguments provided, improved context management
+- **Test Suite Updates**: All tests updated to use async/await consistently
+
+### Documentation Updates
+- Added Document naming rules to CLAUDE.md
+- Added FlowConfig validation rules
+- Added code elegance principles section
+- Updated guide_for_ai.md to API reference format
+
 ## Stability Notice
 
-**Current Version**: 0.1.6
+**Current Version**: 0.1.8
 **Status**: Internal Preview
 **API Stability**: Unstable - Breaking changes expected
 **Recommended Use**: Learning and reference only

ai_pipeline_core-0.1.8/ai_pipeline_core/__init__.py

@@ -0,0 +1,77 @@
+"""Pipeline Core - Shared infrastructure for AI pipelines."""
+
+from . import llm
+from .documents import (
+    Document,
+    DocumentList,
+    FlowDocument,
+    TaskDocument,
+    canonical_name_key,
+    sanitize_url,
+)
+from .flow import FlowConfig, FlowOptions
+from .llm import (
+    AIMessages,
+    AIMessageType,
+    ModelName,
+    ModelOptions,
+    ModelResponse,
+    StructuredModelResponse,
+)
+from .logging import (
+    LoggerMixin,
+    LoggingConfig,
+    StructuredLoggerMixin,
+    get_pipeline_logger,
+    setup_logging,
+)
+from .logging import get_pipeline_logger as get_logger
+from .pipeline import pipeline_flow, pipeline_task
+from .prefect import disable_run_logger, prefect_test_harness
+from .prompt_manager import PromptManager
+from .settings import settings
+from .tracing import TraceInfo, TraceLevel, trace
+
+__version__ = "0.1.8"
+
+__all__ = [
+    # Config/Settings
+    "settings",
+    # Logging
+    "get_logger",
+    "get_pipeline_logger",
+    "LoggerMixin",
+    "LoggingConfig",
+    "setup_logging",
+    "StructuredLoggerMixin",
+    # Documents
+    "Document",
+    "DocumentList",
+    "FlowDocument",
+    "TaskDocument",
+    "canonical_name_key",
+    "sanitize_url",
+    # Flow/Task
+    "FlowConfig",
+    "FlowOptions",
+    # Pipeline decorators (with tracing)
+    "pipeline_task",
+    "pipeline_flow",
+    # Prefect decorators (clean, no tracing)
+    "prefect_test_harness",
+    "disable_run_logger",
+    # LLM
+    "llm",
+    "ModelName",
+    "ModelOptions",
+    "ModelResponse",
+    "StructuredModelResponse",
+    "AIMessages",
+    "AIMessageType",
+    # Tracing
+    "trace",
+    "TraceLevel",
+    "TraceInfo",
+    # Utils
+    "PromptManager",
+]

{ai_pipeline_core-0.1.6 → ai_pipeline_core-0.1.8}/ai_pipeline_core/documents/__init__.py

@@ -2,10 +2,13 @@ from .document import Document
 from .document_list import DocumentList
 from .flow_document import FlowDocument
 from .task_document import TaskDocument
+from .utils import canonical_name_key, sanitize_url
 
 __all__ = [
     "Document",
     "DocumentList",
     "FlowDocument",
     "TaskDocument",
+    "canonical_name_key",
+    "sanitize_url",
 ]

{ai_pipeline_core-0.1.6 → ai_pipeline_core-0.1.8}/ai_pipeline_core/documents/document.py

@@ -26,12 +26,27 @@ TModel = TypeVar("TModel", bound=BaseModel)
 
 
 class Document(BaseModel, ABC):
-    """Abstract base class for all documents"""
+    """Abstract base class for all documents.
+
+    Warning: Document subclasses should NOT start with 'Test' prefix as this
+    causes conflicts with pytest test discovery. Classes with 'Test' prefix
+    will be rejected at definition time.
+    """
 
     MAX_CONTENT_SIZE: ClassVar[int] = 25 * 1024 * 1024  # 25MB default
     DESCRIPTION_EXTENSION: ClassVar[str] = ".description.md"
     MARKDOWN_LIST_SEPARATOR: ClassVar[str] = "\n\n---\n\n"
 
+    def __init_subclass__(cls, **kwargs: Any) -> None:
+        """Validate subclass names to prevent pytest conflicts."""
+        super().__init_subclass__(**kwargs)
+        if cls.__name__.startswith("Test"):
+            raise TypeError(
+                f"Document subclass '{cls.__name__}' cannot start with 'Test' prefix. "
+                "This causes conflicts with pytest test discovery. "
+                "Please use a different name (e.g., 'SampleDocument', 'ExampleDocument')."
+            )
+
     def __init__(self, **data: Any) -> None:
         """Prevent direct instantiation of abstract Document class."""
         if type(self) is Document:

ai_pipeline_core-0.1.8/ai_pipeline_core/flow/__init__.py

@@ -0,0 +1,7 @@
+from .config import FlowConfig
+from .options import FlowOptions
+
+__all__ = [
+    "FlowConfig",
+    "FlowOptions",
+]

{ai_pipeline_core-0.1.6 → ai_pipeline_core-0.1.8}/ai_pipeline_core/flow/config.py

@@ -14,6 +14,27 @@ class FlowConfig(ABC):
     INPUT_DOCUMENT_TYPES: ClassVar[list[type[FlowDocument]]]
     OUTPUT_DOCUMENT_TYPE: ClassVar[type[FlowDocument]]
 
+    def __init_subclass__(cls, **kwargs):
+        """Validate that OUTPUT_DOCUMENT_TYPE is not in INPUT_DOCUMENT_TYPES."""
+        super().__init_subclass__(**kwargs)
+
+        # Skip validation for the abstract base class itself
+        if cls.__name__ == "FlowConfig":
+            return
+
+        # Ensure required attributes are defined
+        if not hasattr(cls, "INPUT_DOCUMENT_TYPES"):
+            raise TypeError(f"FlowConfig {cls.__name__} must define INPUT_DOCUMENT_TYPES")
+        if not hasattr(cls, "OUTPUT_DOCUMENT_TYPE"):
+            raise TypeError(f"FlowConfig {cls.__name__} must define OUTPUT_DOCUMENT_TYPE")
+
+        # Validate that output type is not in input types
+        if cls.OUTPUT_DOCUMENT_TYPE in cls.INPUT_DOCUMENT_TYPES:
+            raise TypeError(
+                f"FlowConfig {cls.__name__}: OUTPUT_DOCUMENT_TYPE "
+                f"({cls.OUTPUT_DOCUMENT_TYPE.__name__}) cannot be in INPUT_DOCUMENT_TYPES"
+            )
+
     @classmethod
     def get_input_document_types(cls) -> list[type[FlowDocument]]:
         """

ai_pipeline_core-0.1.8/ai_pipeline_core/flow/options.py

@@ -0,0 +1,26 @@
+from typing import TypeVar
+
+from pydantic import Field
+from pydantic_settings import BaseSettings, SettingsConfigDict
+
+from ai_pipeline_core.llm import ModelName
+
+T = TypeVar("T", bound="FlowOptions")
+
+
+class FlowOptions(BaseSettings):
+    """Base configuration for AI Pipeline flows."""
+
+    core_model: ModelName | str = Field(
+        default="gpt-5",
+        description="Primary model for complex analysis and generation tasks.",
+    )
+    small_model: ModelName | str = Field(
+        default="gpt-5-mini",
+        description="Fast, cost-effective model for simple tasks and orchestration.",
+    )
+
+    model_config = SettingsConfigDict(frozen=True, extra="ignore")
+
+
+__all__ = ["FlowOptions"]

{ai_pipeline_core-0.1.6 → ai_pipeline_core-0.1.8}/ai_pipeline_core/llm/client.py

@@ -118,11 +118,13 @@ async def _generate_with_retry(
                 span.set_attributes(response.get_laminar_metadata())
                 Laminar.set_span_output(response.content)
                 if not response.content:
-                    # disable cache in case of empty response
-                    completion_kwargs["extra_body"]["cache"] = {"no-cache": True}
                     raise ValueError(f"Model {model} returned an empty response.")
                 return response
         except (asyncio.TimeoutError, ValueError, Exception) as e:
+            if not isinstance(e, asyncio.TimeoutError):
+                # disable cache if it's not a timeout because it may cause an error
+                completion_kwargs["extra_body"]["cache"] = {"no-cache": True}
+
             logger.warning(
                 "LLM generation failed (attempt %d/%d): %s",
                 attempt + 1,
@@ -167,7 +169,7 @@ T = TypeVar("T", bound=BaseModel)
 
 @trace(ignore_inputs=["context"])
 async def generate_structured(
-    model: ModelName,
+    model: ModelName | str,
     response_format: type[T],
     *,
     context: AIMessages = AIMessages(),

{ai_pipeline_core-0.1.6 → ai_pipeline_core-0.1.8}/ai_pipeline_core/llm/model_options.py

@@ -4,6 +4,7 @@ from pydantic import BaseModel
 
 
 class ModelOptions(BaseModel):
+    temperature: float | None = None
     system_prompt: str | None = None
     search_context_size: Literal["low", "medium", "high"] | None = None
     reasoning_effort: Literal["low", "medium", "high"] | None = None
@@ -21,6 +22,9 @@ class ModelOptions(BaseModel):
             "extra_body": {},
         }
 
+        if self.temperature:
+            kwargs["temperature"] = self.temperature
+
         if self.max_completion_tokens:
             kwargs["max_completion_tokens"] = self.max_completion_tokens