pandas>=1.5.0

# chunk op: free, no extra deps.

# classify op mode='zero_shot' (default), embed op provider='sentence_transformers' (default):
#   pip install transformers torch sentence-transformers

# classify op mode='llm', embed op provider='litellm':
#   pip install litellm

# embed op provider='openai':
#   pip install openai
# embed op provider='cohere':
#   pip install cohere

# write_vector_store op provider='chromadb' (default):
#   pip install chromadb
# write_vector_store op provider='pinecone':
#   pip install pinecone
# write_vector_store op provider='qdrant':
#   pip install qdrant-client
# write_vector_store op provider='s3vectors':
#   pip install boto3
# write_vector_store op provider='lancedb':
#   pip install lancedb

# embed op reuse_from.provider='qdrant' / 'pinecone' / 'lancedb' (reuse_from.provider='chromadb'
# needs nothing beyond the chromadb install above):
#   pip install qdrant-client
#   pip install pinecone
#   pip install lancedb

# sensitivity op engine='presidio':
#   pip install presidio-analyzer presidio-anonymizer spacy
#   python -m spacy download en_core_web_sm

# source: {kind: warehouse_query, resource_key: ...} against DuckDB (via a registered Dagster resource):
#   pip install dagster-duckdb duckdb
# source: {kind: warehouse_query, database_url_env_var: ...} (bare SQLAlchemy, no Dagster resource):
#   pip install duckdb duckdb-engine   # or the SQLAlchemy dialect for your warehouse (psycopg2, snowflake-sqlalchemy, etc.)
# execution_mode='sql': no extra pip deps -- SQL runs server-side via the
# warehouse's own resource (snowflake_resource / bigquery_resource /
# databricks resource), which must already be registered.
