Retrieval Configuration

Configure how your document chunks are retrieved and ranked

All changes saved

Quick Configuration Presets

Retrieval Method

Vector Search

Pure semantic similarity search using embeddings

Speed95%
Accuracy85%
Cost30%

Pros:

Fast retrieval
Semantic understanding
Language agnostic

Cons:

May miss exact matches
Requires good embeddings

Keyword Search (BM25)

Traditional keyword-based search with BM25 ranking

Speed98%
Accuracy75%
Cost10%

Pros:

Exact match capability
Low cost
No embeddings needed

Cons:

No semantic understanding
Language dependent

Hybrid Search

RECOMMENDED

Combines vector and keyword search for best results

Speed92%
Accuracy95%
Cost40%

Pros:

Best of both worlds
Highly accurate
Flexible weighting

Cons:

Higher complexity
More compute required

Graph-Based Retrieval

Uses knowledge graphs for relationship-aware search

Speed70%
Accuracy90%
Cost60%

Pros:

Relationship aware
Multi-hop reasoning
Context preservation

Cons:

Complex setup
Higher latency
Requires graph DB

HyDE

Hypothetical Document Embeddings for improved retrieval

Speed80%
Accuracy92%
Cost50%

Pros:

Better query understanding
Improved relevance
Context expansion

Cons:

Additional LLM calls
Higher latency
Cost overhead

Multi-Query Retrieval

Generates multiple query variations for comprehensive search

Speed75%
Accuracy93%
Cost45%

Pros:

Comprehensive coverage
Query robustness
Better recall

Cons:

Multiple searches
Higher compute
Result merging complexity