Types of RAG
Open any method to explore its agent or see its architecture. Together they form Ontology-guided Adaptive RAG: a router picks the right retrieval for each question.
Ontology-guided Hybrid RAG
Vector, lexical search, and reranking together for balanced enterprise retrieval.
Structured RAG + semantic layer
Authoritative answers from governed enterprise data and queries.
Agentic RAG
An agent plans the search and uses tools. Powerful, but more resource-intensive.
Ontology-guided GraphRAG
Follows relationships between entities, so connected questions get complete answers.
Multimodal RAG
Search across scans, diagrams, tables, audio and video, with citations to page, region or timestamp.
Memory RAG
Governed memory for long-running workflows and prior decisions.
Pick the method by the question and the data
Two check marks: recommended. One: works. Every method cites its sources, declines when the evidence is not enough, and applies access rules inside the search.
| Use case | Hybrid | Structured | Agentic | GraphRAG | Multimodal | Memory | Vector |
|---|---|---|---|---|---|---|---|
| Policies, procedures and contracts for one department (for example HR only)e.g. How many days do we have to dispute a charge? | ✓✓ | – | ✓ | – | ✓ | – | ✓ |
| Learning content: courses, training decks, recorded sessionse.g. Which module covers data retention, and what is the quiz pass mark? | ✓✓ | – | ✓ | ✓ | ✓✓ | ✓ | ✓ |
| Exact codes, IDs, product names and error codese.g. What does error E-6032 mean? | ✓✓ | ✓ | ✓ | ✓ | ✓ | – | – |
| Questions worded differently from the documentse.g. My phone is three years old, can I get a new one? | ✓✓ | – | ✓ | – | ✓ | – | ✓✓ |
| Totals, counts, averages and rankings over tablese.g. Which carrier has the highest total charges? | ✓ | ✓✓ | ✓✓ | – | – | – | ✓ |
| One record looked up exactlye.g. What was the total billed for Rogers in January 2026? | ✓ | ✓✓ | ✓ | ✓ | – | – | ✓ |
| Live data from source systems (changes during the day)e.g. Change a product description in the source, ask again, see the new value | – | ✓✓ | ✓✓ | – | – | – | – |
| How things connect: owners, approvers, dependencies, impacte.g. Which lines are affected if carrier X has an outage? | – | ✓ | ✓ | ✓✓ | – | – | – |
| Enterprise-wide: HR, finance, supply chain and more, with one assistante.g. Which departments are over their software budget and who owns those licenses? | ✓ | ✓ | ✓✓ | ✓✓ | – | ✓ | – |
| Scans, slides, charts, diagrams and photose.g. How many new lines were requested on service order SO-88213? | ✓ | – | ✓ | – | ✓✓ | – | ✓ |
| One question that needs two sources (a document and a table)e.g. Which carrier has the highest total charges, and what is the Zone 2 daily pass price? | ✓ | – | ✓✓ | ✓ | – | – | ✓ |
| Long-running cases, prior decisions and preferencese.g. What did we decide about the Rogers credit dispute last week? | – | – | ✓ | – | – | ✓✓ | – |
| High volume, lowest cost and fastest responsee.g. Self-service FAQ for thousands of employees | ✓ | ✓✓ | – | – | – | ✓ | ✓✓ |
Common Capabilities / Modules
Kept as modules inside the accelerators above, not as standalone accelerators.
RAPTOR-Style Hierarchical RAG
Layered summaries for long documents and broad questions.
Temporal / Event-Aware Retrieval
Answers time-based questions: as-of, during, between and what changed.
Ontology / Semantic Guidance Plug-In
Business glossary, semantic models or ontology used to guide retrieval where available.
Standard Vector RAG
Semantic search over documents. The cheapest and simplest approach.
What's included
Context Engineering Methods ship as a connected set of components you can adopt together or one at a time.
Enterprise RAG and GraphRAG Data Pipeline Accelerator
The ingest, parse, chunk, embed, and extract pipeline that builds the knowledge graph and vector store, then grounds AI responses in trusted enterprise knowledge with hybrid RAG and GraphRAG retrieval. Demo-ready.
Context66 KnowledgeAgent
A grounded question-answering agent over your enterprise knowledge graph. It answers only from what it can retrieve and trace, cites its sources, and hands off cleanly when it does not know.
Reference Architecture for Building Context for AI
The full 8-layer map of every capability behind this engine, from raw source data to the human using it, color-coded live by what's actually built today.
Ready to engineer your context?
See how Context66 grounds your AI in knowledge you can trust. We will walk your team through a live accelerator.

