rag retrieval strategies

Persistent knowledge graphs are an evolution of RAG, addressing its statelessness limitation. This is a direct extension improving retrieval efficiency.

Episodic and semantic memory types use vector stores (Pinecone, Weaviate, Chroma) with semantic similarity retrieval—core RAG pattern. Article maps retrieval methods to memory types.

RAG results are explicitly mentioned as one component of context engineering that must be balanced against other information sources.

Semantic memory retrieval pattern is functionally equivalent to RAG: external knowledge stored separately, retrieved based on query/task, injected into prompt.

The pattern of on-demand context loading based on task state is conceptually identical to RAG: retrieve relevant context when needed, not everything upfront. The research validates this approach for s

Pre-computed knowledge graph + unified indexing is an advanced RAG pattern. Glean's approach is RAG applied at enterprise scale with relationship awareness.

Mentions dynamic context retrieval using 'user query and metadata' to identify what to fetch—aligns with RAG pattern but applied within agent context management.

Relevance ranking is a retrieval strategy optimization; this quantifies its impact in agent memory context.

The Intercom attack vector shows RAG failure mode: customer support agent retrieves data based on instructions embedded in retrieved context. This is a retrieval poisoning problem.

Explicitly mentions RAG and vectorless RAG as components of agentic AI stack, positioning them as data grounding layer

RAG retrieval is listed as one of the components managed within the context lifecycle in context engineering

RAG integration is mentioned as core topic; RAG is fundamentally about context retrieval and prioritization

Similar problem to RAG: selecting relevant context from a large pool. Here it's selecting relevant trajectory reasoning; in RAG it's selecting relevant documents. Both use attention signals.

Retrieved documents (Layer 3) are positioned as a distinct context layer; this implies RAG as a context engineering problem, not just a search problem.

MCP resources interface is a retrieval mechanism for structured data access. Unlike traditional RAG (vector similarity), MCP enables exact retrieval + semantic understanding of what data exists. This

Unlocking context from unstructured data (documents, audio, video, chat) relates to RAG implementation strategy for agent context enrichment.

Supacrawl is a specialized form of RAG for structured data: instead of embedding text documents, it creates a queryable database snapshot agents can explore. The pattern of 'preserving structured cont

RAG is a specific implementation of context engineering—managing what knowledge the model accesses at decision time.

MCP servers can implement RAG-like retrieval patterns. The protocol standardizes how retrieval systems expose context to LLMs.

LlamaIndex is a primary RAG framework; context engineering relates directly to what context gets retrieved and how it's prioritized.

Retrieving relevant context from long-term memory is fundamentally a retrieval problem, essential for RAG effectiveness

Context 7 and similar servers implement retrieval of live documentation/data—functionally similar to RAG but using protocol-based access instead of traditional vector search.

Servers like Firecrawl, Exa, NotebookLM are retrieval mechanisms that implement RAG-like patterns at the protocol level.

MCP can be viewed as a protocol layer above RAG—standardizing how agents retrieve and request context from knowledge sources.

The Reflector role (evaluating what context stays) aligns with relevance scoring in RAG systems, but at the composition level rather than retrieval level.

MCP provides architectural foundation for how retrieval systems expose relevant context to AI—moves beyond prompt-level RAG to protocol-level context exposure

The failure mode 'using wrong or irrelevant cases' is a RAG failure—retrieving documents that don't actually answer the query or contradict other retrieved documents. The solution requires smarter ret

Offloading and compaction strategies are essentially RAG-like retrieval patterns applied within agent context management.

External knowledge as a context source directly relates to RAG patterns and retrieval strategy design

MCP servers can provide the retrieval layer for RAG systems. The protocol standardizes how to surface retrieved context to AI models.

Article explicitly mentions 'retrieve the most relevant' information, identifying retrieval as critical component of context engineering architecture

ArtifactFS is a specialized RAG pattern: retrieve critical path (file tree), fetch detailed content on-demand. Priority-based retrieval.

references/ folders function as context retrieval mechanism—agents read 'on demand' rather than loading all context upfront, similar to retrieval-augmented generation pattern.

Structured work primitives become the source documents for RAG retrieval. Better retrieval quality because the structured data has semantic meaning (not just log text).

Managing 'procedural knowledge and memory' at enterprise scale requires retrieval/recall strategies similar to RAG but for operational context

The inference of missing context—agents determining what additional information they need—mirrors RAG systems determining when and what to retrieve. The benchmark tests this capability.

The examples (access to funding announcements, rate agreements, PI history) describe retrieval strategy design—ensuring right information is available when needed.

The partitioning + tiered memory architecture is a retrieval strategy: organize data so that relevant subset can be quickly surfaced. This is architecturally similar to RAG chunking and indexing strat

Compaction and offloading strategies map to retrieval/compression patterns in RAG systems

Brave Search MCP server is a form of real-time retrieval—extending Claude's knowledge beyond training cutoff, similar to RAG pattern but via standardized protocol.

Better token management often means better retrieval strategies (RAG) to avoid redundant context. Cost constraints incentivize efficient information retrieval rather than dumping all context into ever

AGENTS.md generation and retrieval mimics RAG pattern—compress and index context (repo guidance) for efficient agent access.

Just-in-time data fetching vs upfront metadata dumps is fundamentally a retrieval strategy question—when and how to fetch context for agent use.

LlamaIndex integration with document-heavy RAG pipelines suggests memory layer must support semantic + structured retrieval—moving beyond pure semantic search.

MCP servers effectively implement RAG infrastructure at the protocol level, giving models standardized way to retrieve external knowledge without custom SDK integration.

Distributing context through shared repos and registries, evaluating context quality, observing how agents use it—these align with RAG system patterns and retrieval evaluation practices.

Demonstrates 5.7% improvement on retrieval-augmented generation through better context collaboration—shows RAG can be improved via orchestration, not just retrieval quality.

Context package composition and type libraries are methodologically related to what information to retrieve and how to structure it for AI consumption.

MCP servers can expose retrieval capabilities (database access, file system, APIs). This is one mechanism for implementing RAG-like patterns.

Cron jobs updating documentation/skill files function similarly to RAG refresh cycles—ensuring agents have fresh, relevant context to retrieve from.