Definition
Knowledge Refresh is the process of updating the underlying vector store or knowledge base in a RAG pipeline to ensure retrieved context reflects the most current data. This involves managing document upserts, identifying delta changes in source content, and purging deprecated entries to maintain high retrieval precision.
Updates the external data the agent 'looks up,' rather than retraining or fine-tuning the LLM's internal weights.
"A Newsstand Refresh: Swapping out yesterday's newspapers for today's editions so the reader never cites outdated headlines."
- Incremental Indexing(Component)
- Stale Context(Problem Solved)
- Vector Database(Prerequisite)
- Data Sync Latency(Architectural Trade-off)
Conceptual Overview
Knowledge Refresh is the process of updating the underlying vector store or knowledge base in a RAG pipeline to ensure retrieved context reflects the most current data. This involves managing document upserts, identifying delta changes in source content, and purging deprecated entries to maintain high retrieval precision.
Disambiguation
Updates the external data the agent 'looks up,' rather than retraining or fine-tuning the LLM's internal weights.
Visual Analog
A Newsstand Refresh: Swapping out yesterday's newspapers for today's editions so the reader never cites outdated headlines.