markdown-flow 0.2.19__py3-none-any.whl → 0.2.31__py3-none-any.whl

markdown_flow/providers/openai.py

@@ -0,0 +1,369 @@
+"""
+OpenAI-Compatible Provider Implementation
+
+Provides a production-ready OpenAI-compatible LLM provider with debug mode,
+token tracking, and comprehensive metadata.
+"""
+
+import time
+from collections.abc import Generator
+from typing import Any
+
+from ..llm import LLMProvider
+from .config import ProviderConfig
+
+
+try:
+    from openai import OpenAI
+except ImportError:
+    OpenAI = None  # type: ignore[misc, assignment]
+
+
+class OpenAIProvider(LLMProvider):
+    """
+    OpenAI-compatible LLM provider implementation.
+
+    Features:
+    - Debug mode with colorized console output
+    - Automatic token usage tracking
+    - Comprehensive metadata (model, temperature, processing time, tokens, timestamp)
+    - Instance-level model/temperature override support
+    - Streaming and non-streaming modes
+    """
+
+    def __init__(self, config: ProviderConfig):
+        """
+        Initialize OpenAI provider.
+
+        Args:
+            config: Provider configuration
+
+        Raises:
+            ImportError: If openai package is not installed
+            ValueError: If configuration is invalid
+        """
+        if OpenAI is None:
+            raise ImportError("The 'openai' package is required for OpenAIProvider. Install it with: pip install openai")
+
+        self.config = config
+        self.client = OpenAI(
+            api_key=config.api_key,
+            base_url=config.base_url,
+            timeout=config.timeout,
+        )
+        self._last_metadata: dict[str, Any] = {}
+
+    def complete(
+        self,
+        messages: list[dict[str, str]],
+        model: str | None = None,
+        temperature: float | None = None,
+    ) -> str:
+        """
+        Non-streaming LLM call.
+
+        Args:
+            messages: Message list
+            model: Optional model override
+            temperature: Optional temperature override
+
+        Returns:
+            LLM response content
+
+        Raises:
+            Exception: If API call fails
+        """
+        # Determine actual model and temperature (instance override > provider default)
+        actual_model = model if model is not None else self.config.model
+        actual_temperature = temperature if temperature is not None else self.config.temperature
+
+        # Debug output: Request info
+        if self.config.debug:
+            self._print_request_info(messages, actual_model, actual_temperature)
+
+        # Format messages
+        formatted_messages = self._format_messages(messages)
+
+        # Record start time
+        start_time = time.time()
+
+        try:
+            # Make API call
+            response = self.client.chat.completions.create(
+                model=actual_model,
+                messages=formatted_messages,
+                temperature=actual_temperature,
+            )
+
+            # Calculate processing time
+            processing_time_ms = int((time.time() - start_time) * 1000)
+
+            # Extract content
+            if not response.choices or len(response.choices) == 0:
+                raise Exception("API response error: no choices returned")
+
+            choice = response.choices[0]
+            if not choice.message:
+                raise Exception("Response has no message field")
+
+            content = choice.message.content or ""
+
+            # Extract token usage
+            usage = response.usage
+            metadata = {
+                "model": actual_model,
+                "temperature": actual_temperature,
+                "provider": "openai-compatible",
+                "processing_time": processing_time_ms,
+                "timestamp": int(time.time()),
+            }
+
+            if usage:
+                metadata.update(
+                    {
+                        "prompt_tokens": usage.prompt_tokens,
+                        "output_tokens": usage.completion_tokens,
+                        "total_tokens": usage.total_tokens,
+                    }
+                )
+
+            # Save metadata for retrieval by MarkdownFlow
+            self._last_metadata = metadata
+
+            # Debug output: Response metadata
+            if self.config.debug:
+                self._print_response_metadata(metadata)
+
+            return content
+
+        except Exception as e:
+            raise Exception(f"API request failed: {str(e)}") from e
+
+    def stream(
+        self,
+        messages: list[dict[str, str]],
+        model: str | None = None,
+        temperature: float | None = None,
+    ) -> Generator[str, None, None]:
+        """
+        Streaming LLM call.
+
+        Args:
+            messages: Message list
+            model: Optional model override
+            temperature: Optional temperature override
+
+        Yields:
+            Incremental LLM response content
+
+        Raises:
+            Exception: If API call fails
+        """
+        # Determine actual model and temperature
+        actual_model = model if model is not None else self.config.model
+        actual_temperature = temperature if temperature is not None else self.config.temperature
+
+        # Debug output: Request info
+        if self.config.debug:
+            self._print_request_info(messages, actual_model, actual_temperature)
+
+        # Format messages
+        formatted_messages = self._format_messages(messages)
+
+        # Record start time
+        start_time = time.time()
+
+        try:
+            # Create streaming response
+            stream = self.client.chat.completions.create(
+                model=actual_model,
+                messages=formatted_messages,
+                temperature=actual_temperature,
+                stream=True,
+            )
+
+            for chunk in stream:
+                if chunk.choices and chunk.choices[0].delta.content:
+                    yield chunk.choices[0].delta.content
+
+            # Calculate processing time after stream completes
+            processing_time_ms = int((time.time() - start_time) * 1000)
+
+            # Save metadata for retrieval by MarkdownFlow
+            metadata = {
+                "model": actual_model,
+                "temperature": actual_temperature,
+                "provider": "openai-compatible",
+                "processing_time": processing_time_ms,
+                "timestamp": int(time.time()),
+                "stream_done": True,
+            }
+            self._last_metadata = metadata
+
+            # Debug output: Stream completion info
+            if self.config.debug:
+                self._print_response_metadata(metadata)
+
+        except Exception as e:
+            raise ValueError(f"Streaming request failed: {str(e)}") from e
+
+    def get_last_metadata(self) -> dict[str, Any]:
+        """
+        Get metadata from the last LLM call.
+
+        This method allows MarkdownFlow to retrieve comprehensive metadata including
+        token usage, processing time, and other information from the most recent
+        complete() or stream() call.
+
+        Returns:
+            Dictionary containing metadata:
+            - model: Model name used
+            - temperature: Temperature value used
+            - provider: Provider identifier
+            - processing_time: Processing time in milliseconds
+            - timestamp: Unix timestamp
+            - prompt_tokens: Number of input tokens (if available)
+            - output_tokens: Number of output tokens (if available)
+            - total_tokens: Total tokens (if available)
+            - stream_done: True if this was a completed stream (stream mode only)
+
+        Example:
+            >>> provider = create_default_provider()
+            >>> content = provider.complete(messages)
+            >>> metadata = provider.get_last_metadata()
+            >>> print(f"Used {metadata['total_tokens']} tokens")
+        """
+        return self._last_metadata.copy()
+
+    def _format_messages(self, messages: list[dict[str, str]]) -> list[dict[str, str]]:
+        """
+        Format messages for API call.
+
+        Args:
+            messages: Raw message list
+
+        Returns:
+            Formatted message list
+        """
+        formatted = []
+        for msg in messages:
+            if isinstance(msg, dict) and "role" in msg and "content" in msg:
+                formatted.append(
+                    {
+                        "role": msg["role"],
+                        "content": str(msg["content"]),
+                    }
+                )
+            else:
+                # Fallback for non-standard format
+                formatted.append(
+                    {
+                        "role": "user",
+                        "content": str(msg),
+                    }
+                )
+        return formatted
+
+    def _print_request_info(self, messages: list[dict[str, str]], model: str, temperature: float) -> None:
+        """
+        Print colorized request information to console (debug mode).
+
+        Args:
+            messages: Message list
+            model: Model name
+            temperature: Temperature value
+        """
+        print("\033[97m\033[44m[ ====== LLM Request Start ====== ]\033[0m")
+        print(f"\033[30m\033[42mmodel\033[0m: {model}")
+        print(f"\033[30m\033[42mtemperature\033[0m: {temperature}")
+
+        for message in messages:
+            role = message.get("role", "user")
+            content = message.get("content", "")
+            # Truncate long content for readability
+            display_content = content
+            print(f"\033[30m\033[43m{role}\033[0m: {display_content}")
+
+        print("\033[97m\033[44m[ ====== LLM Request End ====== ]\033[0m")
+
+    def _print_response_metadata(self, metadata: dict[str, Any]) -> None:
+        """
+        Print colorized response metadata to console (debug mode).
+
+        Args:
+            metadata: Response metadata dictionary
+        """
+        print("\033[97m\033[42m[ ====== LLM Response Metadata ====== ]\033[0m")
+
+        # Essential fields
+        print(f"\033[36mmodel:\033[0m {metadata.get('model', 'N/A')}")
+        print(f"\033[36mtemperature:\033[0m {metadata.get('temperature', 'N/A')}")
+        print(f"\033[36mprovider:\033[0m {metadata.get('provider', 'N/A')}")
+        print(f"\033[36mprocessing_time:\033[0m {metadata.get('processing_time', 'N/A')} ms")
+
+        # Token usage (if available)
+        if "prompt_tokens" in metadata:
+            print(
+                f"\033[36mprompt_tokens:\033[0m \033[33m{metadata['prompt_tokens']}\033[0m  "
+                f"\033[36moutput_tokens:\033[0m \033[33m{metadata['output_tokens']}\033[0m  "
+                f"\033[36mtotal_tokens:\033[0m \033[32m{metadata['total_tokens']}\033[0m"
+            )
+
+        print(f"\033[36mtimestamp:\033[0m {metadata.get('timestamp', 'N/A')}")
+
+        if metadata.get("stream_done"):
+            print("\033[36mstream:\033[0m completed")
+
+        print("\033[97m\033[42m[ ====== ======================= ====== ]\033[0m")
+
+
+def create_provider(config: ProviderConfig | None = None) -> OpenAIProvider:
+    """
+    Create an OpenAI provider instance.
+
+    Args:
+        config: Optional provider configuration. If None, uses default config
+                (reads from environment variables).
+
+    Returns:
+        OpenAIProvider instance
+
+    Raises:
+        ValueError: If configuration is invalid
+        ImportError: If openai package is not installed
+
+    Example:
+        >>> config = ProviderConfig(api_key="sk-...", model="gpt-4")
+        >>> provider = create_provider(config)
+    """
+    if config is None:
+        config = ProviderConfig()
+    return OpenAIProvider(config)
+
+
+def create_default_provider() -> OpenAIProvider:
+    """
+    Create an OpenAI provider with default configuration.
+
+    Reads configuration from environment variables:
+    - LLM_API_KEY: API key (required)
+    - LLM_BASE_URL: Base URL (default: https://api.openai.com/v1)
+    - LLM_MODEL: Model name (default: gpt-3.5-turbo)
+    - LLM_TEMPERATURE: Temperature (default: 0.7)
+    - LLM_DEBUG: Debug mode (default: false)
+    - LLM_TIMEOUT: Request timeout in seconds (default: None, no timeout)
+
+    Returns:
+        OpenAIProvider instance with default config
+
+    Raises:
+        ValueError: If LLM_API_KEY is not set
+        ImportError: If openai package is not installed
+
+    Example:
+        >>> # Set environment variable first
+        >>> import os
+        >>> os.environ["LLM_API_KEY"] = "sk-..."
+        >>> provider = create_default_provider()
+    """
+    return create_provider()

markdown_flow/utils.py

@@ -19,6 +19,7 @@ from .constants import (
     COMPILED_PERCENT_VARIABLE_REGEX,
     COMPILED_PRESERVE_FENCE_REGEX,
     COMPILED_SINGLE_PIPE_SPLIT_REGEX,
+    CONTEXT_BUTTON_OPTIONS_TEMPLATE,
     CONTEXT_CONVERSATION_TEMPLATE,
     CONTEXT_QUESTION_MARKER,
     CONTEXT_QUESTION_TEMPLATE,
@@ -67,7 +68,7 @@ def is_preserved_content_block(content: str) -> bool:
     Check if content is completely preserved content block.
 
     Preserved blocks are entirely wrapped by markers with no external content.
-    Supports inline (===content===) and multiline (!=== ... !===) formats.
+    Supports inline (===content===), multiline (!=== ... !===) formats, and mixed formats.
 
     Args:
         content: Content to check
@@ -81,61 +82,50 @@ def is_preserved_content_block(content: str) -> bool:
 
     lines = content.split("\n")
 
-    # Check if all non-empty lines are inline format (!===content!===)
-    all_inline_format = True
-    has_any_content = False
-
-    for line in lines:
-        stripped_line = line.strip()
-        if stripped_line:  # Non-empty line
-            has_any_content = True
-            # Check if inline format: ===content===
-            match = COMPILED_INLINE_PRESERVE_REGEX.match(stripped_line)
-            if match:
-                # Ensure inner content exists and contains no ===
-                inner_content = match.group(1).strip()
-                if not inner_content or "===" in inner_content:
-                    all_inline_format = False
-                    break
-            else:
-                all_inline_format = False  # type: ignore[unreachable]
-                break
-
-    # If all lines are inline format, return directly
-    if has_any_content and all_inline_format:
-        return True
-
-    # Check multiline format using state machine
+    # Use state machine to validate that all non-empty content is preserved
     state = "OUTSIDE"  # States: OUTSIDE, INSIDE
-    has_content_outside = False  # Has external content
-    has_preserve_blocks = False  # Has preserve blocks
+    has_preserve_content = False
 
     for line in lines:
         stripped_line = line.strip()
 
+        # Check if this line is a fence marker (!===)
         if COMPILED_PRESERVE_FENCE_REGEX.match(stripped_line):
             if state == "OUTSIDE":
                 # Enter preserve block
                 state = "INSIDE"
-                has_preserve_blocks = True
+                has_preserve_content = True
             elif state == "INSIDE":
                 # Exit preserve block
                 state = "OUTSIDE"
-            # !=== lines don't count as external content
-        else:
-            # Non-!=== lines
-            if stripped_line:  # type: ignore[unreachable]  # Non-empty line
-                if state == "OUTSIDE":
-                    # External content found
-                    has_content_outside = True
-                    break
-                # Internal content doesn't affect judgment
+            # Fence markers themselves are valid preserved content
+            continue
+
+        # Non-fence lines
+        if stripped_line:  # Non-empty line
+            if state == "INSIDE":
+                # Inside fence block, this is valid preserved content
+                has_preserve_content = True
+            else:
+                # Outside fence block, check if it's inline format
+                match = COMPILED_INLINE_PRESERVE_REGEX.match(stripped_line)
+                if match:
+                    # Ensure inner content exists and contains no ===
+                    inner_content = match.group(1).strip()
+                    if inner_content and "===" not in inner_content:
+                        # Valid inline format
+                        has_preserve_content = True
+                    else:
+                        # Invalid inline format
+                        return False
+                else:
+                    # Not fence, not inline format -> external content
+                    return False
 
     # Judgment conditions:
-    # 1. Must have preserve blocks
-    # 2. Cannot have external content
-    # 3. Final state must be OUTSIDE (all blocks closed)
-    return has_preserve_blocks and not has_content_outside and state == "OUTSIDE"
+    # 1. Must have preserved content
+    # 2. Final state must be OUTSIDE (all fence blocks closed)
+    return has_preserve_content and state == "OUTSIDE"
 
 
 def extract_interaction_question(content: str) -> str | None:
@@ -479,6 +469,7 @@ def generate_smart_validation_template(
     target_variable: str,
     context: list[dict[str, Any]] | None = None,
     interaction_question: str | None = None,
+    buttons: list[dict[str, str]] | None = None,
 ) -> str:
     """
     Generate smart validation template based on context and question.
@@ -487,19 +478,28 @@ def generate_smart_validation_template(
         target_variable: Target variable name
         context: Context message list with role and content fields
         interaction_question: Question text from interaction block
+        buttons: Button options list with display and value fields
 
     Returns:
         Generated validation template
     """
     # Build context information
     context_info = ""
-    if interaction_question or context:
+    if interaction_question or context or buttons:
         context_parts = []
 
         # Add question information (most important, put first)
         if interaction_question:
             context_parts.append(CONTEXT_QUESTION_TEMPLATE.format(question=interaction_question))
 
+        # Add button options information
+        if buttons:
+            button_displays = [btn.get("display", "") for btn in buttons if btn.get("display")]
+            if button_displays:
+                button_options_str = ", ".join(button_displays)
+                button_info = CONTEXT_BUTTON_OPTIONS_TEMPLATE.format(button_options=button_options_str)
+                context_parts.append(button_info)
+
         # Add conversation context
         if context:
             for msg in context: