Text Index Stores

Text Index Stores are Heta's full-text search index interface. They handle:

chunk id -> searchable text + metadata

IndexFullText writes to TextIndexStoreProtocol; full_text_search queries it. Text index stores do not save raw files, chunk JSON, vectors, or graph facts.

Implementations

Store	Use
InMemoryTextIndexStore	Local development, unit tests, and lightweight demos.
ElasticsearchTextIndexStore	Production full-text retrieval with Elasticsearch BM25 ranking.

Elasticsearch

Install:

pip install "heta-framework[elasticsearch]"

Configure:

from heta_framework.common.stores import (
    ElasticsearchTextIndexStore,
    ElasticsearchTextIndexStoreConfig,
)

text_index = ElasticsearchTextIndexStore(
    ElasticsearchTextIndexStoreConfig(
        hosts="http://localhost:9200",
        request_timeout=30,
    )
)

Use it in a recipe:

recipe = KnowledgeRecipe(
    stores=KnowledgeStores(
        objects=objects,
        text_index=text_index,
    ),
    steps=(
        ParseDocuments(),
        SplitDocuments(),
        IndexFullText(),
    ),
)

Query:

response = await kb.query(
    "flight control fault",
    mode="full_text_search",
    top_k=5,
)

Relationship To SQL Text Search

IndexFullText and PersistChunks are separate steps:

Step	Store	Query mode	Use
IndexFullText	TextIndexStoreProtocol	full_text_search	BM25-style full-text retrieval.
PersistChunks	SQLStoreProtocol	sql_text_search	SQL persistence and lightweight LIKE retrieval.

If you only need Elasticsearch full-text search, use IndexFullText. If you also need SQL evidence tables, chunk tables, or lightweight LIKE retrieval, add PersistChunks.

Scope

Text Index Stores write searchable text, preserve retrieval metadata, return relevance-ranked results, and connect to systems such as Elasticsearch.

They do not store raw files, persist chunk JSON, compute embeddings, run vector search, build graphs, or manage the KnowledgeBase lifecycle.