`llm`¶

Full name: tenets.core.summarizer.llm

llm¶

LLM-based summarization strategies.

This module provides integration with Large Language Models (LLMs) for high-quality summarization. Supports OpenAI, Anthropic, and OpenRouter APIs.

NOTE: These strategies incur API costs. Use with caution and appropriate rate limiting. Always check pricing before using in production.

Classes¶

LLMProvider¶

Bases: Enum

Supported LLM providers.

LLMConfig`dataclass`¶

Python

LLMConfig(provider: LLMProvider = LLMProvider.OPENAI, model: str = 'gpt-4o-mini', api_key: Optional[str] = None, base_url: Optional[str] = None, temperature: float = 0.3, max_tokens: int = 500, system_prompt: str = 'You are an expert at summarizing code and technical documentation. \nYour summaries are concise, accurate, and preserve critical technical details.', user_prompt: str = 'Summarize the following text to approximately {target_percent}% of its original length. \nFocus on the most important information and maintain technical accuracy.\n\nText to summarize:\n{text}\n\nSummary:', retry_attempts: int = 3, retry_delay: float = 1.0, timeout: float = 30.0)

Configuration for LLM summarization.

ATTRIBUTE	DESCRIPTION
`provider`	LLM provider to use TYPE:`LLMProvider`
`model`	Model name/ID TYPE:`str`
`api_key`	API key (if not in environment) TYPE:`Optional[str]`
`base_url`	Base URL for API (for custom endpoints) TYPE:`Optional[str]`
`temperature`	Sampling temperature (0-1) TYPE:`float`
`max_tokens`	Maximum tokens in response TYPE:`int`
`system_prompt`	System prompt template TYPE:`str`
`user_prompt`	User prompt template TYPE:`str`
`retry_attempts`	Number of retry attempts TYPE:`int`
`retry_delay`	Delay between retries in seconds TYPE:`float`
`timeout`	Request timeout in seconds TYPE:`float`

Functions¶

get_api_key¶

Python

get_api_key() -> Optional[str]

Get API key from config or environment.

RETURNS	DESCRIPTION
`Optional[str]`	API key or None

Source code in tenets/core/summarizer/llm.py

Python

def get_api_key(self) -> Optional[str]:
    """Get API key from config or environment.

    Returns:
        API key or None
    """
    if self.api_key:
        return self.api_key

    # Check environment variables
    env_vars = {
        LLMProvider.OPENAI: "OPENAI_API_KEY",
        LLMProvider.ANTHROPIC: "ANTHROPIC_API_KEY",
        LLMProvider.OPENROUTER: "OPENROUTER_API_KEY",
    }

    env_var = env_vars.get(self.provider)
    if env_var:
        return os.getenv(env_var)

    return None

LLMSummarizer¶

Python

LLMSummarizer(config: Optional[LLMConfig] = None)

Base class for LLM-based summarization.

Provides common functionality for different LLM providers. Handles API calls, retries, and error handling.

Initialize LLM summarizer.

PARAMETER	DESCRIPTION
`config`	LLM configuration TYPE:`Optional[LLMConfig]`DEFAULT:`None`

Source code in tenets/core/summarizer/llm.py

Python

def __init__(self, config: Optional[LLMConfig] = None):
    """Initialize LLM summarizer.

    Args:
        config: LLM configuration
    """
    self.config = config or LLMConfig()
    self.logger = get_logger(__name__)
    self.client = None
    self._initialize_client()

Functions¶

summarize¶

Python

summarize(text: str, target_ratio: float = 0.3, max_length: Optional[int] = None, min_length: Optional[int] = None, custom_prompt: Optional[str] = None) -> str

Summarize text using LLM.

PARAMETER	DESCRIPTION
`text`	Text to summarize TYPE:`str`
`target_ratio`	Target compression ratio TYPE:`float`DEFAULT:`0.3`
`max_length`	Maximum summary length TYPE:`Optional[int]`DEFAULT:`None`
`min_length`	Minimum summary length TYPE:`Optional[int]`DEFAULT:`None`
`custom_prompt`	Custom prompt override TYPE:`Optional[str]`DEFAULT:`None`

RETURNS	DESCRIPTION
`str`	Summarized text

RAISES	DESCRIPTION
`RuntimeError`	If API call fails after retries

Source code in tenets/core/summarizer/llm.py

Python

def summarize(
    self,
    text: str,
    target_ratio: float = 0.3,
    max_length: Optional[int] = None,
    min_length: Optional[int] = None,
    custom_prompt: Optional[str] = None,
) -> str:
    """Summarize text using LLM.

    Args:
        text: Text to summarize
        target_ratio: Target compression ratio
        max_length: Maximum summary length
        min_length: Minimum summary length
        custom_prompt: Custom prompt override

    Returns:
        Summarized text

    Raises:
        RuntimeError: If API call fails after retries
    """
    if not self.client:
        raise RuntimeError(f"No client initialized for {self.config.provider.value}")

    # Prepare prompt
    target_percent = int(target_ratio * 100)

    if custom_prompt:
        user_prompt = custom_prompt.format(
            text=text,
            target_percent=target_percent,
            max_length=max_length,
            min_length=min_length,
        )
    else:
        user_prompt = self.config.user_prompt.format(text=text, target_percent=target_percent)

    # Add length constraints to prompt if specified
    if max_length:
        user_prompt += f"\nMaximum length: {max_length} characters"
    if min_length:
        user_prompt += f"\nMinimum length: {min_length} characters"

    # Make API call with retries
    for attempt in range(self.config.retry_attempts):
        try:
            summary = self._call_api(user_prompt)

            # Validate length constraints
            if max_length and len(summary) > max_length:
                summary = summary[:max_length].rsplit(" ", 1)[0] + "..."
            elif min_length and len(summary) < min_length:
                # Request longer summary
                user_prompt += f"\n\nThe summary is too short. Please provide more detail."
                continue

            return summary

        except Exception as e:
            self.logger.warning(
                f"API call failed (attempt {attempt + 1}/{self.config.retry_attempts}): {e}"
            )
            if attempt < self.config.retry_attempts - 1:
                time.sleep(self.config.retry_delay * (2**attempt))  # Exponential backoff
            else:
                raise RuntimeError(
                    f"Failed to summarize after {self.config.retry_attempts} attempts: {e}"
                )

    return text[:max_length] if max_length else text  # Fallback

estimate_cost¶

Python

estimate_cost(text: str) -> Dict[str, float]

Estimate cost of summarization.

PARAMETER	DESCRIPTION
`text`	Text to summarize TYPE:`str`

RETURNS	DESCRIPTION
`Dict[str, float]`	Dictionary with cost estimates

Source code in tenets/core/summarizer/llm.py

Python

def estimate_cost(self, text: str) -> Dict[str, float]:
    """Estimate cost of summarization.

    Args:
        text: Text to summarize

    Returns:
        Dictionary with cost estimates
    """
    # Rough token estimation (1 token ≈ 4 characters)
    input_tokens = len(text) // 4
    output_tokens = int(input_tokens * 0.3)  # Assume 30% compression

    # Pricing per 1K tokens (as of 2024)
    pricing = {
        "gpt-4o-mini": {"input": 0.00015, "output": 0.0006},
        "gpt-4o": {"input": 0.005, "output": 0.015},
        "gpt-4": {"input": 0.03, "output": 0.06},
        "gpt-4-turbo": {"input": 0.01, "output": 0.03},
        "claude-3-opus": {"input": 0.015, "output": 0.075},
        "claude-3-sonnet": {"input": 0.003, "output": 0.015},
        "claude-3-haiku": {"input": 0.00025, "output": 0.00125},
    }

    model_pricing = pricing.get(self.config.model, {"input": 0.001, "output": 0.002})

    input_cost = (input_tokens / 1000) * model_pricing["input"]
    output_cost = (output_tokens / 1000) * model_pricing["output"]
    total_cost = input_cost + output_cost

    return {
        "input_tokens": input_tokens,
        "output_tokens": output_tokens,
        "input_cost": input_cost,
        "output_cost": output_cost,
        "total_cost": total_cost,
        "currency": "USD",
    }

LLMSummaryStrategy¶

Python

LLMSummaryStrategy(provider: Union[str, LLMProvider] = LLMProvider.OPENAI, model: str = 'gpt-4o-mini', api_key: Optional[str] = None)

LLM-based summarization strategy for use with Summarizer.

Wraps LLMSummarizer to match the SummarizationStrategy interface.

WARNING: This strategy incurs API costs. Always estimate costs before use.

Initialize LLM strategy.

PARAMETER	DESCRIPTION
`provider`	LLM provider name or enum TYPE:`Union[str, LLMProvider]`DEFAULT:`OPENAI`
`model`	Model to use TYPE:`str`DEFAULT:`'gpt-4o-mini'`
`api_key`	API key (if not in environment) TYPE:`Optional[str]`DEFAULT:`None`

Source code in tenets/core/summarizer/llm.py

Python

def __init__(
    self,
    provider: Union[str, LLMProvider] = LLMProvider.OPENAI,
    model: str = "gpt-4o-mini",
    api_key: Optional[str] = None,
):
    """Initialize LLM strategy.

    Args:
        provider: LLM provider name or enum
        model: Model to use
        api_key: API key (if not in environment)
    """
    self.logger = get_logger(__name__)

    # Convert string to enum if needed
    if isinstance(provider, str):
        provider = LLMProvider(provider.lower())

    # Create config
    config = LLMConfig(provider=provider, model=model, api_key=api_key)

    # Initialize summarizer
    self.summarizer = LLMSummarizer(config)

    # Warn about costs
    self.logger.warning(
        f"LLM summarization enabled with {provider.value}/{model}. "
        f"This will incur API costs. Use estimate_cost() to check pricing."
    )

Functions¶

summarize¶

Python

summarize(text: str, target_ratio: float = 0.3, max_length: Optional[int] = None, min_length: Optional[int] = None) -> str

Summarize text using LLM.

PARAMETER	DESCRIPTION
`text`	Input text TYPE:`str`
`target_ratio`	Target compression ratio TYPE:`float`DEFAULT:`0.3`
`max_length`	Maximum summary length TYPE:`Optional[int]`DEFAULT:`None`
`min_length`	Minimum summary length TYPE:`Optional[int]`DEFAULT:`None`

RETURNS	DESCRIPTION
`str`	LLM-generated summary

Source code in tenets/core/summarizer/llm.py

Python

def summarize(
    self,
    text: str,
    target_ratio: float = 0.3,
    max_length: Optional[int] = None,
    min_length: Optional[int] = None,
) -> str:
    """Summarize text using LLM.

    Args:
        text: Input text
        target_ratio: Target compression ratio
        max_length: Maximum summary length
        min_length: Minimum summary length

    Returns:
        LLM-generated summary
    """
    return self.summarizer.summarize(
        text, target_ratio=target_ratio, max_length=max_length, min_length=min_length
    )

estimate_cost¶

Python

estimate_cost(text: str) -> Dict[str, float]

Estimate cost for summarizing text.

PARAMETER	DESCRIPTION
`text`	Text to summarize TYPE:`str`

RETURNS	DESCRIPTION
`Dict[str, float]`	Cost estimate dictionary

Source code in tenets/core/summarizer/llm.py

Python

def estimate_cost(self, text: str) -> Dict[str, float]:
    """Estimate cost for summarizing text.

    Args:
        text: Text to summarize

    Returns:
        Cost estimate dictionary
    """
    return self.summarizer.estimate_cost(text)

Functions¶

create_llm_summarizer¶

Python

create_llm_summarizer(provider: str = 'openai', model: Optional[str] = None, api_key: Optional[str] = None) -> LLMSummaryStrategy

Create an LLM summarizer with defaults.

PARAMETER	DESCRIPTION
`provider`	Provider name (openai, anthropic, openrouter) TYPE:`str`DEFAULT:`'openai'`
`model`	Model name (uses provider default if None) TYPE:`Optional[str]`DEFAULT:`None`
`api_key`	API key (uses environment if None) TYPE:`Optional[str]`DEFAULT:`None`

RETURNS	DESCRIPTION
`LLMSummaryStrategy`	Configured LLMSummaryStrategy

summarizer = create_llm_summarizer("openai", "gpt-4o-mini") >>> summary = summarizer.summarize(long_text, target_ratio=0.2)

Source code in tenets/core/summarizer/llm.py

Python

def create_llm_summarizer(
    provider: str = "openai", model: Optional[str] = None, api_key: Optional[str] = None
) -> LLMSummaryStrategy:
    """Create an LLM summarizer with defaults.

    Args:
        provider: Provider name (openai, anthropic, openrouter)
        model: Model name (uses provider default if None)
        api_key: API key (uses environment if None)

    Returns:
        Configured LLMSummaryStrategy

    Example:
    >>> summarizer = create_llm_summarizer("openai", "gpt-4o-mini")
        >>> summary = summarizer.summarize(long_text, target_ratio=0.2)
    """
    # Default models for each provider
    default_models = {
        "openai": "gpt-4o-mini",
        "anthropic": "claude-3-haiku-20240307",
        "openrouter": "openai/gpt-4o-mini",
        "local": "llama2",
    }

    if model is None:
        model = default_models.get(provider.lower(), "gpt-4o-mini")

    return LLMSummaryStrategy(provider=provider, model=model, api_key=api_key)

llm¶

llm¶

Classes¶

LLMProvider¶

LLMConfigdataclass¶

Functions¶

get_api_key¶

LLMSummarizer¶

Functions¶

summarize¶

estimate_cost¶

LLMSummaryStrategy¶

Functions¶

summarize¶

estimate_cost¶

Functions¶

create_llm_summarizer¶

`llm`¶

LLMConfig`dataclass`¶