POMA AI ⬣ Docs

Appearance

Install the integration:

bash
pip install 'poma[qdrant]'

PomaQdrant is a QdrantClient subclass with POMA-specific helpers.

python
import os

from poma import PrimeCut
from poma.integrations.qdrant import PomaQdrant

client = PrimeCut()
result = client.ingest("example.pdf", show_progress=True)

qdrant = PomaQdrant(
    url=os.environ["QDRANT_URL"],
    api_key=os.environ["QDRANT_API_KEY"],
    cloud_inference=True,
    collection_name="poma-docs",
    dense_model="sentence-transformers/all-MiniLM-L6-v2",
    sparse_model="Qdrant/bm25",
    dense_size=384,
    auto_create_collection=True,
)

qdrant.upsert_poma_points(result)

cheatsheets = qdrant.get_cheatsheets(
    query="What does the document say about retention?",
    limit=3,
)

print(cheatsheets[0]["content"])

Accepted input types ​

upsert_poma_points(...) accepts:

  • typed PomaResult
  • legacy chunk-data dictionaries
  • a path to a .poma archive

Collection creation ​

If you set auto_create_collection=True, also set dense_size.

The default sparse_model is "Qdrant/bm25", so the convenience query path uses hybrid dense+sparse retrieval unless you set sparse_model=None.

Query modes ​

get_cheatsheets(...) supports three modes:

  • query=... for a convenience text query
  • query_obj=... and prefetch=... for direct Qdrant query control
  • results=... to convert existing Qdrant results into cheatsheets

If you call get_cheatsheets(results=...), you can also pass chunk_data=... to rebuild cheatsheets from authoritative chunk data instead of relying on stored payload chunk_details.

Advanced helpers ​

For lower-level point generation or result conversion, import the helpers from poma.integrations.qdrant.qdrant_poma_utils.

The main helpers are:

  • prepare_points_from_chunk_data(...)
  • points_from_chunk_data(...)
  • results_to_cheatsheet_inputs(...)
  • cheatsheets_from_results(...)
  • ensure_collection(...)

See the Qdrant integration reference for the full signatures.