API Overview

On this page

Package exports
Configuration
Models
Strategy interfaces (abstract base classes)
Query analysis
Retrieval
Pipeline (filter / rank / compose)
Guard, learning, cache, budget allocation
Main engine
Legacy compatibility (not in __all__, but importable)
API summary table
Navigation

Package exports

Everything below is exported from the package root (context/__init__'s __all__). Click through to the module page for full details.

Configuration

ContextEngineConfig, ContextConfig, QueryAnalyzerConfig, RetrieverConfig, FilterConfig, RankingConfig, ComposerConfig, GuardConfig, CacheConfig, LengthCounter — see config

Models

QueryAnalysis, QueryIntent, QueryComplexity, RetrievalMethod, SourceQuality, ScoredChunk, ContextResult, ConversationTurn, ConversationHistory, chunk_metadata_dict — see models

Strategy interfaces (abstract base classes)

DeduplicationStrategy, RankingStrategy, CompressionStrategy, BudgetAllocationStrategy, QueryAnalyzerStrategy, ChunkFilterStrategy, ChunkRankingStrategy, ContextComposerStrategy, ContextGuardStrategy, HybridRetrieverStrategy — see strategy

Query analysis

RuleBasedQueryAnalyzer, build_query_analyzer — see query_analyzer

Retrieval

BM25Scorer, VectorRetriever, KeywordRetriever, MetadataRetriever, HybridRetriever — see retriever

Pipeline (filter / rank / compose)

char_counter, make_token_counter, ChunkFilter, CompositeRanker, SmartContextComposer — see pipeline

Guard, learning, cache, budget allocation

ContextGuard, AdaptiveLearningSystem, ContextCache, EqualBudgetAllocation, WeightedBudgetAllocation, ComplexityBasedAllocation — see guard_and_learning

Main engine

ContextEngine, ContextManager — see engine

Legacy compatibility (not in `all`, but importable)

validate_chunk, PrefixDeduplication, HashDeduplication, GreedyCompression, ScoreRanking — see legacy_compat

API summary table

API	Type	Module	Description
`ContextEngine`	Class	`engine`	Main pipeline orchestrator — `index(chunks, ...)` and `index_documents(texts, chunker, ...)`
`ContextManager`	Class	`engine`	Backward-compatible subclass of `ContextEngine`
`ContextEngineConfig` / `ContextConfig`	Dataclass	`config`	Root configuration object
`QueryAnalyzerConfig`	Dataclass	`config`	Query analyzer settings
`RetrieverConfig`	Dataclass	`config`	Retriever weights/toggles
`FilterConfig`	Dataclass	`config`	Chunk filtering settings
`RankingConfig`	Dataclass	`config`	Ranking weights
`ComposerConfig`	Dataclass	`config`	Context composition settings
`GuardConfig`	Dataclass	`config`	Guard thresholds/toggles
`CacheConfig`	Dataclass	`config`	LRU cache settings
`QueryAnalysis`	Dataclass	`models`	Result of analyzing one query
`ScoredChunk`	Dataclass	`models`	Unified wrapper around any chunk object
`ContextResult`	Dataclass	`models`	Final output of `ContextEngine.run()`
`ConversationTurn`	Dataclass	`models`	One Q/A turn
`ConversationHistory`	Dataclass	`models`	Multi-turn conversation state
`chunk_metadata_dict`	Function	`models`	Normalizes a chunk's `metadata` (`dict` or `ChunkMetadata` dataclass) into a plain `dict`
`QueryIntent`	Enum	`models`	7-value intent classification
`QueryComplexity`	Enum	`models`	3-value complexity classification
`RetrievalMethod`	Enum	`models`	4-value retrieval method tag
`SourceQuality`	Enum	`models`	4-value source quality tag
`RuleBasedQueryAnalyzer`	Class	`query_analyzer`	Default `QueryAnalyzerStrategy` implementation
`build_query_analyzer`	Function	`query_analyzer`	Factory for query analyzers
`BM25Scorer`	Class	`retriever`	Dependency-free BM25 keyword scorer
`VectorRetriever`	Class	`retriever`	Native `zerda.vector_database`-backed, external-adapter-backed, or in-memory semantic retrieval
`KeywordRetriever`	Class	`retriever`	BM25-based retrieval
`MetadataRetriever`	Class	`retriever`	Pure metadata-filter retrieval
`HybridRetriever`	Class	`retriever`	Combines the above three via RRF + weighted scoring
`ChunkFilter`	Class	`pipeline`	Noise/length/dedup filtering
`CompositeRanker`	Class	`pipeline`	Multi-factor composite scoring and ranking
`SmartContextComposer`	Class	`pipeline`	Template-based context assembly under budget
`ContextTemplate`	Class	`pipeline`	Formatting template (not exported at package root)
`char_counter`	Function	`pipeline`	Character-based `LengthCounter`
`make_token_counter`	Function	`pipeline`	Builds a tokenizer-based `LengthCounter`
`ContextGuard`	Class	`guard_and_learning`	Context validation
`AdaptiveLearningSystem`	Class	`guard_and_learning`	Rolling-window usage stats + budget suggestion
`ContextCache`	Class	`guard_and_learning`	In-memory LRU+TTL cache
`EqualBudgetAllocation`	Class	`guard_and_learning`	Equal split of a multi-query budget
`WeightedBudgetAllocation`	Class	`guard_and_learning`	Weighted split of a multi-query budget
`ComplexityBasedAllocation`	Class	`guard_and_learning`	Complexity-weighted split of a multi-query budget
`validate_chunk`	Function	`legacy_compat`	Raises `TypeError` on malformed legacy chunks
`PrefixDeduplication`	Class	`legacy_compat`	Legacy prefix-based dedup
`HashDeduplication`	Class	`legacy_compat`	Legacy SHA-256 dedup
`ScoreRanking`	Class	`legacy_compat`	Legacy score-sort ranking
`GreedyCompression`	Class	`legacy_compat`	Legacy greedy context compression

config · models · strategy · query_analyzer
retriever · pipeline · guard_and_learning
engine · legacy_compat

Back to docs home · Architecture · Configuration