Index Full Text

IndexFullText writes chunk text into a TextIndexStore and unlocks BM25-style full-text retrieval.

ParsedChunk JSON -> TextIndexStore

After this step succeeds, the KB can use:

full_text_search

Contract

IndexFullText uses:

stores.objects
stores.text_index

Default input artifact:

chunk_keys

Execution flow:

read chunk_keys
  -> load ParsedChunk JSON
  -> write text and metadata into TextIndexStore
  -> declare chunk_full_text_index
  -> unlock full_text_search

Configuration

IndexFullTextConfig(
    index_name="chunks",
    object_store=None,
    text_index_store=None,
    chunk_keys_artifact="chunk_keys",
)

Parameter	Meaning
`index_name`	Full-text index name.
`object_store`	Named ObjectStore. Defaults to `stores.objects`.
`text_index_store`	Named TextIndexStore. Defaults to `stores.text_index`.
`chunk_keys_artifact`	Artifact containing chunk JSON keys.

Requirements

StepRequirements(
    components=frozenset({
        store_ref("objects"),
        store_ref("text_index"),
    }),
    artifacts=frozenset({"chunk_keys"}),
)

Capabilities

StepCapabilities(
    artifacts=frozenset({"index_full_text_result"}),
    queries=frozenset({"full_text_search"}),
    search_assets=(
        SearchAsset(kind="chunk_full_text_index", ...),
    ),
)

Relationship To PersistChunks

IndexFullText and PersistChunks are separate:

Step	Store	Query mode	Use
`IndexFullText`	`TextIndexStoreProtocol`	`full_text_search`	BM25-style retrieval, usually Elasticsearch in production.
`PersistChunks`	`SQLStoreProtocol`	`sql_text_search`	SQL chunk persistence and lightweight LIKE retrieval.

Use IndexFullText when you need a dedicated full-text index. Use PersistChunks when SQL chunk tables are part of your graph, evidence, or operational workflow.