Semantic search over documents. A simple, cost-efficient approach.
- Status
- Next up
- Type
- Baseline within the Hybrid agent
- Best use
- A first, low-cost assistant over plain documents in one department
- Accuracy and completeness
- Acceptable, but can miss exact terms and relationships
- Cost and performance
- Low cost and relatively simple
The simplest start for plain documents, when cost and speed matter most
Best for
- A first assistant over one department's documents
- Questions worded differently from the text ('a new phone' vs 'device upgrade')
- High-volume FAQ where cost and speed matter most
Ask it things like
- My phone is three years old, can I get a new one?
- What is our data retention period for invoices?
- How do I report a lost device while travelling?
Not the right choice for
- Exact codes, IDs and names (use Hybrid)
- Relationships, totals, scans with region citations (GraphRAG, Structured, Multimodal)
Choose it over Hybrid when budget and speed matter more than exact-code questions, and the documents are plain prose.
- Data it reads
- The same documents as Hybrid.
- How it gets the data
- Embeds each passage and the question, finds the closest passages by meaning, answers only from them, and checks every claim.
- How fresh the answer is
- As of the last time the documents were indexed.
- Relative cost
- Lowest for documents
How Standard Vector RAG works
- 1
Chunk
Parse documents into passages while retaining source IDs, permission scope and source locations.
- 2
Embed
Produce passage embeddings with the selected customer-approved embedding model.
- 3
Store
Index embeddings with references, provenance, sensitivity and retrievable policy/ACL information.
- 4
Search
Embed the question and retrieve the most similar authorized passages.
- 5
Answer & Check
Create a grounded response, associate citations and abstain or clarify when support is inadequate.
Vector search is one of the core retrieval mechanisms used within Ontology-Guided GraphRAG.
One router picks the right method for each question
Context66 combines these methods as Ontology-guided Adaptive RAG. Every answer is checked against its evidence, and the system abstains when the evidence is incomplete.
- User question
- Intent and context detection
- Ontology and semantic interpretation
- Query router
Runs within the Hybrid agent
- Evidence fusion and ranking
- Answer written only from retrieved evidence
- Claim-level validation and citation checking

