Selective Context Inclusion

Send only the files, sections, or conversation turns relevant to the current query. Requires a retrieval or routing step to decide what to keep.

How It Works

Before constructing the prompt, a retrieval system ranks available content by relevance to the current task. Only the top-ranked items are included. Relevance can be determined by embedding similarity, keyword matching, or a learned ranker.

What to Select

• specific files or functions instead of the full repository
• relevant conversation turns instead of full history
• sections of a document instead of the whole text
• specific tool outputs instead of all available results

Trade-offs

• Pros: can achieve very high context reduction with high relevance.
• Cons: retrieval quality determines task success; poor selection drops critical context.

See Also

• Signal-per-Token Optimization — selective inclusion is a primary technique for maximizing signal density.
• Subagents as Context Firewalls — a parent constructing a child's brief performs selective inclusion on the child's behalf.
• Hierarchical Memory — semantic memory retrieval is a form of selective inclusion.
• Grep-Result Tool Output — selective inclusion applied to search results.
• Context Isolation via Subagents — narrows by relevance; isolation keeps the material out of the caller's context entirely.
• Progressive Disclosure — makes the inclusion decision structural: material is split into levels the model pulls in as needed.

Tools

• LangChain RetrievalQA — retrieves relevant documents before answering.
• RAGFlow — retrieval-augmented generation with selective document inclusion.