Support metadata-based filtering for delete operations #99

New Issue

2026-02-16T05:16:32-05:00

yindo commented

2026-02-16 05:16:32 -05:00

Originally created by @yukiharada1228 on GitHub (Dec 16, 2025).

Support metadata-based filtering for delete operations

Feature Request

Current Behavior

Currently, the delete() and adelete() methods only support deletion by document IDs:

async def adelete(
    self,
    ids: Optional[list] = None,
    **kwargs: Any,
) -> Optional[bool]:
    """Delete records from the table."""
    if not ids:
        return False
    # ... only deletes by IDs

Reference: langchain_postgres/v2/async_vectorstore.py:400-419

Proposed Enhancement

Add support for metadata-based filtering in delete operations, similar to how similarity_search supports filtering:

# Example usage
await vectorstore.adelete(filter={"source": "documentation"})
await vectorstore.adelete(filter={"$and": [{"category": "obsolete"}, {"year": {"$lt": 2020}}]})

Motivation

Bulk deletions: Users often need to delete groups of documents based on metadata criteria (e.g., all documents from a specific source, time period, or category)
Existing infrastructure: The codebase already has _create_filter_clause() method that supports metadata filtering for search operations
Consistency: Other vector stores (like Chroma, Pinecone) support metadata-based deletion
Practical use cases:
- Remove all documents from a deprecated data source
- Delete documents older than a certain date
- Clean up documents with specific tags or categories

Implementation Suggestion

The adelete() method could be enhanced to accept an optional filter parameter:

async def adelete(
    self,
    ids: Optional[list] = None,
    filter: Optional[dict] = None,
    **kwargs: Any,
) -> Optional[bool]:
    """Delete records from the table.

    Args:
        ids: List of document IDs to delete
        filter: Metadata filter dictionary for bulk deletion
    """
    if not ids and not filter:
        return False

    if filter:
        safe_filter, filter_dict = self._create_filter_clause(filter)
        query = f'DELETE FROM "{self.schema_name}"."{self.table_name}" WHERE {safe_filter}'
        async with self.engine.connect() as conn:
            await conn.execute(text(query), filter_dict)
            await conn.commit()
    else:
        # ... existing ID-based deletion logic

    return True

This would leverage the existing _create_filter_clause() infrastructure and provide a consistent filtering API across the library.

Alternative: Support both IDs and filters

The implementation could also support combining both IDs and filters:

# Delete specific IDs that also match filter criteria
await vectorstore.adelete(
    ids=["id1", "id2", "id3"],
    filter={"status": "archived"}
)

This would provide maximum flexibility for deletion operations.

Originally created by @yukiharada1228 on GitHub (Dec 16, 2025). # Support metadata-based filtering for delete operations ## Feature Request ### Current Behavior Currently, the `delete()` and `adelete()` methods only support deletion by document IDs: ```python async def adelete( self, ids: Optional[list] = None, **kwargs: Any, ) -> Optional[bool]: """Delete records from the table.""" if not ids: return False # ... only deletes by IDs ``` Reference: `langchain_postgres/v2/async_vectorstore.py:400-419` ### Proposed Enhancement Add support for metadata-based filtering in delete operations, similar to how `similarity_search` supports filtering: ```python # Example usage await vectorstore.adelete(filter={"source": "documentation"}) await vectorstore.adelete(filter={"$and": [{"category": "obsolete"}, {"year": {"$lt": 2020}}]}) ``` ### Motivation 1. **Bulk deletions**: Users often need to delete groups of documents based on metadata criteria (e.g., all documents from a specific source, time period, or category) 2. **Existing infrastructure**: The codebase already has `_create_filter_clause()` method that supports metadata filtering for search operations 3. **Consistency**: Other vector stores (like Chroma, Pinecone) support metadata-based deletion 4. **Practical use cases**: - Remove all documents from a deprecated data source - Delete documents older than a certain date - Clean up documents with specific tags or categories ### Implementation Suggestion The `adelete()` method could be enhanced to accept an optional `filter` parameter: ```python async def adelete( self, ids: Optional[list] = None, filter: Optional[dict] = None, **kwargs: Any, ) -> Optional[bool]: """Delete records from the table. Args: ids: List of document IDs to delete filter: Metadata filter dictionary for bulk deletion """ if not ids and not filter: return False if filter: safe_filter, filter_dict = self._create_filter_clause(filter) query = f'DELETE FROM "{self.schema_name}"."{self.table_name}" WHERE {safe_filter}' async with self.engine.connect() as conn: await conn.execute(text(query), filter_dict) await conn.commit() else: # ... existing ID-based deletion logic return True ``` This would leverage the existing `_create_filter_clause()` infrastructure and provide a consistent filtering API across the library. ### Alternative: Support both IDs and filters The implementation could also support combining both IDs and filters: ```python # Delete specific IDs that also match filter criteria await vectorstore.adelete( ids=["id1", "id2", "id3"], filter={"status": "archived"} ) ``` This would provide maximum flexibility for deletion operations.

yindo added the enhancement label 2026-02-16 05:16:32 -05:00

yindo closed this issue

2026-02-16 05:16:32 -05:00

yindo commented

2026-02-16 05:16:33 -05:00

@dishaprakash commented on GitHub (Jan 21, 2026):

@yukiharada1228 Thank you for contributing to the enhancement of the library! Closing this issue as the PR is merged.

@dishaprakash commented on GitHub (Jan 21, 2026): @yukiharada1228 Thank you for contributing to the enhancement of the library! Closing this issue as the PR is merged.

yindo commented

2026-02-16 05:16:33 -05:00

@yukiharada1228 commented on GitHub (Jan 21, 2026):

@dishaprakash
Thanks for reviewing and merging the PR!
Happy to contribute to the library’s enhancement.
Looking forward to contributing more in the future.

@yukiharada1228 commented on GitHub (Jan 21, 2026): @dishaprakash Thanks for reviewing and merging the PR! Happy to contribute to the library’s enhancement. Looking forward to contributing more in the future.

yindo referenced this issue

2026-02-16 05:16:49 -05:00

[PR #99] [MERGED] address issue for mmr calculation #162

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: langchain-ai/langchain-postgres#99