Teaching an AI agent to forget a deleted Redshift row

Описание и применение

Подключение Cognee к Amazon Redshift позволяет удалять устаревшие строки из графа знаний, предотвращая ошибки ИИ.

An AI agent retained knowledge of a deleted Redshift row, leading to incorrect responses.

The connector successfully removes deleted Redshift rows from Cognee’s knowledge graph, preventing AI agents from referencing stale data.

Configure Cognee with a Redshift source specifying tables or queries.
Run Cognee’s remember() to ingest rows as documents and create summary pages from table comments.
On subsequent runs, Cognee queries only changed rows and retrieves the list of primary keys.
If a key present in the previous run is missing now, Cognee deletes the associated node from the graph.
Safety rules: failed queries abort deletion; late rows with old timestamps are still captured.
Use the Data API for secure, tunnel‑free access to Redshift Serverless.
Test the connector against fakes, then a real Redshift Serverless workgroup.
Verify that deleted rows are removed and summaries include column comments.

An AI agent reads your warehouse. Someone deletes a customer. A week later the agent is still talking about that customer.

Deleted rows are forgotten.

The second sync reported rows: 3, deleted: 3. The deleted ticket's customer vanished from both the documents and the graph.

My first run failed with Redshift endpoint doesn't exist in this region.