Documentation Index
Fetch the complete documentation index at: https://docs.praison.ai/llms.txt
Use this file to discover all available pages before exploring further.
Retrieval Strategies
The retrieval strategy system automatically selects the optimal retrieval approach based on corpus size, ensuring efficient and accurate knowledge retrieval.
Overview
Available strategies:
- DIRECT - Load all content directly (small corpora)
- BASIC - Semantic search only
- HYBRID - Keyword + semantic search
- RERANKED - Hybrid + reranking
- COMPRESSED - Reranked + compression
- HIERARCHICAL - Multi-level summaries for very large corpora
Quick Start
from praisonaiagents.rag import select_strategy, RetrievalStrategy
# Auto-select based on corpus size
strategy = select_strategy(corpus_tokens=5000)
print(f"Selected strategy: {strategy.value}")
# Manual strategy selection
strategy = RetrievalStrategy.HYBRID
Strategy Selection
Automatic Selection
The system automatically selects strategies based on corpus size:
| Corpus Size | Strategy | Description |
|---|
| < 500 tokens | DIRECT | Load all content |
| < 5,000 tokens | BASIC | Semantic search |
| < 20,000 tokens | HYBRID | Keyword + semantic |
| < 50,000 tokens | RERANKED | Hybrid + reranking |
| < 100,000 tokens | COMPRESSED | With compression |
| > 100,000 tokens | HIERARCHICAL | Multi-level summaries |
from praisonaiagents.rag import select_strategy
# Examples
select_strategy(corpus_tokens=100) # -> DIRECT
select_strategy(corpus_tokens=3000) # -> BASIC
select_strategy(corpus_tokens=10000) # -> HYBRID
select_strategy(corpus_tokens=30000) # -> RERANKED
select_strategy(corpus_tokens=75000) # -> COMPRESSED
select_strategy(corpus_tokens=200000) # -> HIERARCHICAL
Manual Override
from praisonaiagents.rag import RetrievalConfig, RetrievalStrategy
config = RetrievalConfig(
strategy=RetrievalStrategy.RERANKED,
top_k=10,
min_score=0.5,
)
Strategy Details
DIRECT Strategy
Best for very small corpora where all content fits in context:
# All content loaded directly
# No retrieval overhead
# Maximum context utilization
BASIC Strategy
Semantic search using embeddings:
from praisonaiagents.rag import SmartRetriever
retriever = SmartRetriever()
results = retriever.search(
query="What is the API key?",
top_k=5,
)
HYBRID Strategy
Combines keyword (BM25) and semantic search:
from praisonaiagents.rag import SmartRetriever
retriever = SmartRetriever(
use_hybrid=True,
keyword_weight=0.3,
semantic_weight=0.7,
)
results = retriever.search(query, top_k=10)
RERANKED Strategy
Adds reranking for improved relevance:
from praisonaiagents.rag import SmartRetriever
retriever = SmartRetriever(use_reranking=True)
results = retriever.search(query, top_k=20)
reranked = retriever.rerank(query, results, top_k=5)
COMPRESSED Strategy
Includes context compression:
from praisonaiagents.rag import ContextCompressor
compressor = ContextCompressor(
max_tokens=4000,
target_ratio=0.5,
)
compressed = compressor.compress(chunks, query=query)
HIERARCHICAL Strategy
Uses multi-level summaries:
from praisonaiagents.rag import HierarchicalSummarizer
summarizer = HierarchicalSummarizer(max_levels=3)
result = summarizer.build_hierarchy(files, base_path="./docs")
answer = summarizer.query("What are the main features?")
CLI Usage
# Search with specific strategy
praisonai knowledge search "query" --strategy hybrid
# Search with reranking
praisonai knowledge search "query" --rerank
# Search with compression
praisonai knowledge search "query" --compress --max-context-tokens 4000
# Combined options
praisonai knowledge search "query" --strategy reranked --rerank --compress
Integration with Agents
from praisonaiagents import Agent
agent = Agent(
name="SmartRetriever",
instructions="Answer questions using the knowledge base.",
knowledge={
"sources": ["./docs"],
"retrieval_k": 10,
"rerank": True,
}
)
response = agent.chat("What are the key features?")
Best Practices
- Start with auto-selection - Let the system choose based on corpus size
- Monitor performance - Use profiling to identify bottlenecks
- Adjust for quality - Increase top_k and enable reranking for better results
- Consider latency - More complex strategies add latency
API Reference
RetrievalStrategy Enum
class RetrievalStrategy(str, Enum):
DIRECT = "direct"
BASIC = "basic"
HYBRID = "hybrid"
RERANKED = "reranked"
COMPRESSED = "compressed"
HIERARCHICAL = "hierarchical"
select_strategy Function
def select_strategy(
corpus_tokens: int = 0,
corpus_files: int = 0,
query_complexity: str = "medium",
) -> RetrievalStrategy:
"""Select optimal strategy based on corpus characteristics."""
