API Reference

Kerb provides a comprehensive set of modules for building LLM applications. Each module is designed to be lightweight, modular, and easy to integrate into your existing projects.

Modules:

• Core Module
  • Import Structure:
  • Document
  • Message
  • ChainStrategy
  • CompressionStrategy
  • ChunkingStrategy
• Agent Module
  • Import Structure:
  • Key Components:
  • Agent
  • run_agent()
  • AgentResult
  • Tool
  • create_tool()
• Cache Module
  • CacheEntry
  • CacheStats
  • BaseCache
  • MemoryCache
  • DiskCache
  • TieredCache
  • LLMCache
  • generate_cache_key()
  • generate_prompt_key()
  • generate_embedding_key()
  • cached()
  • create_memory_cache()
  • create_disk_cache()
  • create_tiered_cache()
  • create_llm_cache()
  • invalidate_expired_entries()
  • export_cache_stats()
  • estimate_cache_size()
• Chunk Module
  • Chunker
  • chunk_text()
  • RecursiveChunker
  • SentenceChunker
  • SemanticChunker
  • CodeChunker
  • MarkdownChunker
  • simple_chunker()
  • overlap_chunker()
  • paragraph_chunker()
  • sliding_window_chunker()
  • token_based_chunker()
  • recursive_chunker()
  • sentence_window_chunker()
  • merge_chunks()
  • optimize_chunk_size()
  • custom_chunker()
• Config Module
  • Config
  • ConfigManager
  • load_config()
  • save_config()
  • ConfigSource
  • ProviderType
  • AppConfig
  • ModelConfig
  • ProviderConfig
  • create_config_manager()
  • create_model_config()
  • create_provider_config()
  • get_openai_config()
  • get_anthropic_config()
  • get_google_config()
  • validate_credentials()
  • load_config_from_file()
  • save_config_to_file()
  • get_default_openai_config()
  • get_default_anthropic_config()
  • get_default_google_config()
  • validate_provider_credentials()
• Context Module
  • ContextItem
  • ContextWindow
  • CompressionResult
  • TruncationStrategy
  • CompressionMethod
  • create_context_window()
  • truncate_context_window()
  • compress_context()
  • auto_compress_window()
  • create_sliding_window()
  • create_token_sliding_window()
  • create_adaptive_window()
  • deduplicate_context()
  • reorder_context()
  • merge_context_windows()
  • optimize_context_for_query()
  • format_context_window()
  • context_to_messages()
  • extract_context_summary()
  • assign_priorities()
  • priority_by_recency()
  • priority_by_relevance()
  • priority_by_diversity()
• Document Module
  • Document
  • DocumentFormat
  • load_document()
  • load_text()
  • load_markdown()
  • load_json()
  • load_csv()
  • load_xml()
  • load_html()
  • load_pdf()
  • load_docx()
  • detect_format()
  • is_supported_format()
  • load_directory()
  • load_from_url()
  • load_from_url_async()
  • merge_documents()
  • extract_text_from_html()
  • strip_markdown()
  • split_into_sentences()
  • split_into_paragraphs()
  • clean_text()
  • remove_extra_newlines()
  • preprocess_pdf_text()
  • preprocess_html_text()
  • preprocess_markdown()
  • extract_metadata()
  • extract_document_stats()
  • extract_urls()
  • extract_emails()
  • extract_dates()
  • extract_phone_numbers()
• Embedding Module
  • EmbeddingModel
  • ModelBackend
  • embed()
  • embed_batch()
  • embed_async()
  • embed_batch_async()
  • embed_batch_stream()
  • embed_batch_stream_async()
  • LocalEmbedder
  • OpenAIEmbedder
  • SentenceTransformerEmbedder
  • local_embed()
  • openai_embed()
  • openai_embed_batch()
  • openai_embed_async()
  • openai_embed_batch_async()
  • sentence_transformer_embed()
  • sentence_transformer_embed_batch()
  • cosine_similarity()
  • euclidean_distance()
  • manhattan_distance()
  • dot_product()
  • batch_similarity()
  • top_k_similar()
  • normalize_vector()
  • vector_magnitude()
  • mean_pooling()
  • weighted_mean_pooling()
  • max_pooling()
  • embedding_dimension()
  • pairwise_similarities()
  • cluster_embeddings()
• Evaluation Module
  • EvaluationResult
  • ComparisonResult
  • BenchmarkResult
  • TestCase
  • EvaluationMetric
  • JudgmentCriterion
  • calculate_bleu()
  • calculate_rouge()
  • calculate_meteor()
  • calculate_exact_match()
  • calculate_f1_score()
  • calculate_semantic_similarity()
  • assess_coherence()
  • assess_fluency()
  • assess_faithfulness()
  • assess_answer_relevance()
  • detect_hallucination()
  • llm_as_judge()
  • pairwise_comparison()
  • ab_test()
  • compare_outputs()
  • run_benchmark()
  • benchmark_prompts()
  • calculate_statistics()
  • confidence_interval()
• Fine-Tuning Module
  • FineTuningProvider
  • DatasetFormat
  • SplitStrategy
  • ValidationLevel
  • TrainingExample
  • TrainingDataset
  • ValidationResult
  • DatasetStats
  • TrainingConfig
  • prepare_dataset()
  • split_dataset()
  • balance_dataset()
  • deduplicate_dataset()
  • sample_dataset()
  • filter_dataset()
  • to_openai_format()
  • to_anthropic_format()
  • from_csv()
  • from_json()
  • write_jsonl()
  • read_jsonl()
  • validate_dataset()
  • estimate_cost()
  • analyze_dataset()
  • check_data_quality()
  • create_training_config()
  • estimate_training_time()
• Generation Module
  • generate()
  • generate_stream()
  • generate_batch()
  • generate_async()
  • Generator
  • GenerationConfig
  • GenerationResponse
  • StreamChunk
  • Usage
  • LLMProvider
  • ModelName
  • MessageRole
  • Message
  • retry_with_exponential_backoff()
  • async_retry_with_exponential_backoff()
  • parse_json_response()
  • validate_response()
  • format_messages()
  • calculate_cost()
  • RateLimiter
  • ResponseCache
  • CostTracker
  • get_cost_summary()
  • reset_cost_tracking()
  • BaseProvider
  • register_provider()
  • get_provider()
  • list_providers()
  • OpenAIGenerator
  • AnthropicGenerator
  • GoogleGenerator
• Memory Module
  • ConversationBuffer
  • Entity
  • ConversationSummary
  • create_sliding_window()
  • create_token_limited_window()
  • create_alternating_window()
  • create_progressive_summary()
  • summarize_conversation()
  • create_hierarchical_summary()
  • save_conversation()
  • load_conversation()
• Multimodal Module
  • MediaType
  • ImageFormat
  • AudioFormat
  • VideoFormat
  • VisionModel
  • TranscriptionModel
  • EmbeddingModelMultimodal
  • ImageInfo
  • AudioInfo
  • VideoInfo
  • TranscriptionResult
  • VisionAnalysis
  • MultiModalContent
  • detect_media_type()
  • get_mime_type()
  • validate_media_file()
  • calculate_file_checksum()
  • load_image()
  • get_image_info()
  • convert_image_format()
  • image_to_base64()
  • base64_to_image()
  • extract_dominant_colors()
  • calculate_image_hash()
  • analyze_image_with_vision_model()
  • embed_multimodal()
  • compute_multimodal_similarity()
  • get_audio_info()
  • convert_audio_format()
  • transcribe_audio()
  • transcribe_audio_async()
  • extract_audio_from_video()
  • get_video_info()
  • extract_video_frames()
  • create_video_thumbnail()
  • build_multimodal_prompt()
  • build_anthropic_multimodal_content()
  • build_google_multimodal_content()
• Parsing Module
  • ParseMode
  • ValidationLevel
  • ParseResult
  • ValidationResult
  • extract_json()
  • parse_json()
  • fix_json()
  • extract_json_array()
  • extract_json_object()
  • ensure_json_output()
  • ensure_list_output()
  • ensure_dict_output()
  • validate_json_schema()
  • parse_to_pydantic()
  • pydantic_to_schema()
  • validate_pydantic()
  • pydantic_to_function()
  • format_function_call()
  • format_tool_call()
  • parse_function_call()
  • format_function_result()
  • extract_code_blocks()
  • extract_xml_tag()
  • extract_markdown_sections()
  • extract_list_items()
  • parse_markdown_table()
  • validate_output()
  • retry_parse_with_fixes()
  • clean_llm_output()
• Preprocessing Module
  • NormalizationLevel
  • LanguageDetectionMode
  • DeduplicationMode
  • ContentType
  • LanguageResult
  • QualityMetrics
  • NormalizationConfig
  • normalize_text()
  • normalize_whitespace()
  • normalize_unicode()
  • normalize_quotes()
  • normalize_dashes()
  • remove_accents()
  • clean_html()
  • clean_markdown()
  • remove_urls()
  • remove_emails()
  • remove_phone_numbers()
  • remove_special_chars()
  • remove_extra_whitespace()
  • remove_control_chars()
  • strip_punctuation()
  • normalize_case()
  • to_title_case()
  • to_sentence_case()
  • preserve_acronyms()
  • detect_language()
  • detect_language_batch()
  • is_language()
  • filter_by_language()
  • get_supported_languages()
  • deduplicate_exact()
  • deduplicate_fuzzy()
  • deduplicate_semantic()
  • deduplicate_lines()
  • deduplicate_sentences()
  • find_duplicates()
  • compute_text_hash()
  • filter_by_length()
  • filter_by_pattern()
  • filter_profanity()
  • filter_pii()
  • detect_spam()
  • filter_by_quality()
  • filter_non_ascii()
  • classify_content_type()
  • detect_code()
  • detect_sentiment()
  • measure_readability()
  • count_words()
  • count_sentences()
  • count_paragraphs()
  • expand_contractions()
  • standardize_numbers()
  • standardize_dates()
  • extract_entities()
  • segment_sentences()
  • segment_words()
  • preprocess_batch()
  • preprocess_pipeline()
  • truncate_text()
  • split_long_text()
• Prompt Module
  • render_template()
  • render_template_safe()
  • validate_template()
  • extract_template_variables()
  • PromptVersion
  • PromptRegistry
  • create_version()
  • register_prompt()
  • get_prompt()
  • list_versions()
  • compare_versions()
  • select_version()
  • FewShotExample
  • ExampleSelector
  • create_example()
  • select_examples()
  • format_examples()
  • compress_prompt()
  • optimize_whitespace()
  • analyze_prompt()
• Retrieval Module
  • Document
  • SearchResult
  • HybridSearchConfig
  • FilterConfig
  • rewrite_query()
  • expand_query()
  • generate_sub_queries()
  • keyword_search()
  • semantic_search()
  • hybrid_search()
  • rerank_results()
  • reciprocal_rank_fusion()
  • diversify_results()
  • compress_context()
  • filter_results()
  • format_results()
  • results_to_context()
• Safety Module
  • SafetyLevel
  • ContentCategory
  • PIIType
  • ToxicityLevel
  • SafetyResult
  • PIIMatch
  • ModerationResult
  • Guardrail
  • moderate_content()
  • check_toxicity()
  • check_sexual_content()
  • check_violence()
  • check_hate_speech()
  • check_self_harm()
  • check_profanity()
  • detect_pii()
  • redact_pii()
  • detect_email()
  • detect_phone()
  • detect_ssn()
  • detect_credit_card()
  • detect_ip_address()
  • detect_url()
  • anonymize_text()
  • detect_prompt_injection()
  • detect_jailbreak()
  • detect_system_prompt_leak()
  • detect_role_confusion()
  • check_input_safety()
  • validate_output()
  • filter_output()
  • check_output_safety()
  • ensure_safe_json()
  • detect_code_injection()
  • create_guardrail()
  • apply_guardrails()
  • check_content_policy()
  • validate_against_rules()
  • sanitize_input()
  • escape_special_chars()
  • validate_url_safety()
  • check_file_upload()
  • detect_data_exfiltration()
  • match_patterns()
  • classify_content()
  • score_content()
  • extract_entities()
• Testing Module
  • MockBehavior
  • FixtureFormat
  • MockResponse
  • TestCase
  • TestResult
  • FixtureData
  • PromptFixture
  • ResponseFixture
  • PromptTestCase
  • SnapshotData
  • PerformanceMetrics
  • CostReport
  • MockLLM
  • MockStreamingLLM
  • create_mock_llm()
  • FixtureManager
  • load_fixtures()
  • save_fixtures()
  • TestDataset
  • create_dataset()
  • load_dataset()
  • assert_response_contains()
  • assert_response_json()
  • assert_response_quality()
• Tokenizer Module
  • count_tokens()
  • batch_count_tokens()
  • count_tokens_for_messages()
  • truncate_to_token_limit()
  • Tokenizer
  • tokens_to_chars()
  • chars_to_tokens()
  • estimate_cost()
  • optimize_token_usage()

Module Overview

Core

Shared types and interfaces used across all modules.

Agent

Agent orchestration and execution patterns for multi-step reasoning and autonomous task completion.

Cache

Response and embedding caching mechanisms to reduce API costs and improve latency.

Chunk

Text chunking utilities for optimal context window usage and retrieval performance.

Config

Configuration management for models, providers, API keys, and application settings.

Context

Context window management and token budget tracking for LLM conversations.

Document

Document loading and processing utilities for PDFs, web pages, DOCX, and more.

Embedding

Embedding generation with support for multiple providers and similarity search helpers.

Evaluation

Metrics and benchmarking tools for evaluating LLM outputs (BLEU, ROUGE, BERTScore, etc.).

Fine-Tuning

Model fine-tuning utilities and large dataset preparation for training custom models.

Generation

Unified LLM text generation with multi-provider support (OpenAI, Anthropic, Gemini, Cohere).

Memory

Conversation memory and entity tracking for building stateful applications.

Multimodal

Image, audio, and video processing utilities for multimodal LLM applications.

Parsing

Output parsing and validation for JSON, structured data, and function calls.

Preprocessing

Text cleaning, normalization, and preprocessing utilities for LLM inputs.

Prompt

Prompt engineering utilities, templates, and chain-of-thought patterns.

Retrieval

RAG (Retrieval-Augmented Generation) and vector search utilities for semantic retrieval.

Safety

Content moderation, safety filters, and input validation.

Testing

Testing utilities and helpers for LLM outputs and evaluation workflows.

Tokenizer

Token counting and text splitting utilities compatible with any model.