LlamaIndex `LlamaIndexRetrievalOperator`¶

Load a persisted LlamaIndex index and run similarity search. Designed to sit between LlamaIndexEmbeddingOperator (which builds the index) and LLMOperator (which synthesises an answer from the retrieved chunks).

Passes the embedding model directly to load_index_from_storage(..., embed_model=...) – no LlamaIndex Settings mutation. The embedding model must match the one used when the index was originally built.

Basic usage¶

airflow/providers/common/ai/example_dags/example_llamaindex_hook.py[source]

@dag(schedule=None, tags=["example"])
def example_llamaindex_retrieve():
    """Load a persisted index and run similarity search."""

    retrieve = LlamaIndexRetrievalOperator(
        task_id="retrieve",
        query="{{ params.query }}",
        index_persist_dir="/opt/airflow/data/library_index",
        embed_model="text-embedding-3-small",
        llm_conn_id="llamaindex_default",
        top_k=5,
    )

    retrieve

query is templated, so DAG-run params, XCom, and Variables all flow through cleanly.

Cloud-persisted indexes¶

index_persist_dir accepts the same local-path-or-URI shape as LlamaIndexEmbeddingOperator.persist_dir. Pass persist_conn_id to point at the Airflow connection that holds cloud credentials. The operator raises FileNotFoundError with a clear “did you run LlamaIndexEmbeddingOperator first?” message when the path is missing.

Bring-your-own embedding model¶

Same shape as LlamaIndexEmbeddingOperator: embed_model accepts either a string model name (OpenAI via the hook) or a pre-built BaseEmbedding instance for non-OpenAI vendors. See the BYO example in LlamaIndex LlamaIndexEmbeddingOperator.

Parameters¶

Parameter	Description
`query`	The query string. Templated.
`index_persist_dir`	Local path or storage URI pointing at the persisted index. Templated.
`persist_conn_id`	Cloud credentials connection ID for `index_persist_dir` URIs. Templated.
`embed_model`	String model name OR pre-built `BaseEmbedding` instance. Must match the model used when the index was built. Templated.
`llm_conn_id`	Airflow connection ID used when `embed_model` is a string. Falls back to `LlamaIndexHook.default_conn_name` (`llamaindex_default`) when `None`.
`embed_conn_id`	Optional separate connection ID for the embedding provider. Falls back to `llm_conn_id` when `None`.
`top_k`	Number of top similarity results to return (default 5).

Output¶

Returns a dict with:

{
    "query": str,
    "chunks": [
        {
            "text": str,
            "score": float,
            "metadata": dict,
            "node_id": str,
        },
        ...
    ],
}

LlamaIndex LlamaIndexRetrievalOperator¶