API Reference¶
Complete API documentation for COSMIC.
Core Classes¶
| Class | Description |
|---|---|
| COSMICChunker | Main chunking orchestrator |
| Document | Input document representation |
| COSMICChunk | Output chunk with metadata |
| COSMICConfig | Configuration system |
Quick Reference¶
Basic Usage¶
from cosmic import COSMICChunker, Document
# Create chunker
chunker = COSMICChunker()
# Create document
doc = Document.from_text("Your text here...")
# Chunk document
chunks = chunker.chunk_document(doc, strategy="auto")
With Configuration¶
from cosmic import COSMICChunker, COSMICConfig
from cosmic.core.config import DCSConfig
config = COSMICConfig(
dcs=DCSConfig(alpha=0.5, beta=0.3, gamma=0.2)
)
chunker = COSMICChunker(config=config)
Modules¶
cosmic¶
Main package exports:
from cosmic import (
COSMICChunker,
COSMICChunk,
COSMICConfig,
Document,
ProcessingMode,
Intent,
DomainType,
)
cosmic.core¶
Core data structures:
cosmic.core.chunk- COSMICChunk dataclasscosmic.core.config- Configuration classescosmic.core.document- Document classcosmic.core.enums- Enumerations
cosmic.pipeline¶
Pipeline stages:
cosmic.pipeline.structure- Stage 1: Structure Analysiscosmic.pipeline.semantic- Stage 2: Semantic Boundariescosmic.pipeline.domain- Stage 3: Domain Classificationcosmic.pipeline.fusion- Stage 4: Boundary Fusioncosmic.pipeline.verification- Stage 5: LLM Verificationcosmic.pipeline.reference- Stage 6: Reference Linking
cosmic.scoring¶
Scoring algorithms:
cosmic.scoring.dcs- Discourse Coherence Scorercosmic.scoring.clustering- MST Clustering
cosmic.models¶
ML model wrappers:
cosmic.models.embeddings- Sentence-transformerscosmic.models.llm- LLM clientcosmic.models.ollama- Ollama integrationcosmic.models.coreference- spaCy coreference
cosmic.fallback¶
Fallback strategies:
cosmic.fallback.semantic_only- Semantic-only chunkingcosmic.fallback.sliding_window- Sliding windowcosmic.fallback.fixed_length- Fixed-length splitting
Enumerations¶
ProcessingMode¶
from cosmic import ProcessingMode
ProcessingMode.FULL_COSMIC # Full 6-stage pipeline
ProcessingMode.SEMANTIC_ONLY # DCS-based boundaries
ProcessingMode.SLIDING_WINDOW # Similarity-based
ProcessingMode.FIXED_LENGTH # Token-based
Intent¶
from cosmic import Intent
Intent.DEFINE # Defines a concept
Intent.EXPLAIN # Explains how something works
Intent.LIST # Enumerates items
Intent.ARGUE # Makes an argument
Intent.DESCRIBE # Describes characteristics
Intent.INSTRUCT # Provides instructions
Intent.SUMMARIZE # Summarizes information
Intent.COMPARE # Compares alternatives
Intent.NARRATE # Tells a story
DomainType¶
from cosmic import DomainType
DomainType.TECHNICAL
DomainType.MEDICAL
DomainType.LEGAL
DomainType.SCIENTIFIC
DomainType.GENERAL