Retrieval Strategies

PraisonAI picks the best retrieval strategy from your indexed corpus size — from direct loading for tiny folders to hierarchical search for massive libraries.

from praisonaiagents import Agent

agent = Agent(
    name="Researcher",
    instructions="Answer from the knowledge base.",
    knowledge={"sources": ["docs/"], "retrieval_k": 10, "rerank": True},
)
response = agent.start("What are the key features?")

The user asks a question; PraisonAI selects the best retrieval strategy for the indexed corpus size.

How It Works

Quick Start

Simple Usage

Strategy selection is automatic when you attach knowledge to an agent:

from praisonaiagents import Agent

agent = Agent(
    name="Researcher",
    instructions="Answer from the knowledge base.",
    knowledge={"sources": ["docs/"], "retrieval_k": 10, "rerank": True},
)
response = agent.start("What are the key features?")

With Configuration

Override the auto-selected strategy via KnowledgeConfig.config:

from praisonaiagents import Agent, KnowledgeConfig

agent = Agent(
    name="Researcher",
    instructions="Answer from documents.",
    knowledge=KnowledgeConfig(
        sources=["docs/"],
        rerank=True,
        config={"strategy": "hybrid", "top_k": 10, "hybrid": True},
    ),
)

Strategy Selection

File count	Strategy	Technique
< 10	DIRECT	Load all content
< 100	BASIC	Semantic search
< 1,000	HYBRID	Keyword + semantic
< 10,000	RERANKED	Hybrid + reranking
< 100,000	COMPRESSED	Reranked + compression
≥ 100,000	HIERARCHICAL	Multi-level summaries

from praisonaiagents.rag import select_strategy
from praisonaiagents import CorpusStats

stats = CorpusStats(file_count=500)
strategy = select_strategy(stats)
print(strategy.value)  # hybrid

Configuration Options

Option	Type	Default	Description
`strategy`	`str`	`"auto"`	`auto`, `direct`, `basic`, `hybrid`, `reranked`, `compressed`, `hierarchical`
`top_k`	`int`	`5`	Chunks to retrieve
`rerank`	`bool`	`False`	Enable cross-encoder reranking
`hybrid`	`bool`	`False`	Combine keyword and semantic search
`compress`	`bool`	`False`	Compress retrieved context

CLI Usage

praisonai knowledge search "query" --strategy hybrid
praisonai knowledge search "query" --rerank --compress --max-context-tokens 4000

Best Practices

Start with auto-selection

Let the SDK pick the strategy from corpus size unless you have measured latency or quality requirements.

Enable rerank for large corpora

Set rerank=True when retrieval_k is high — reranking improves relevance on noisy result sets.

Use hybrid for mixed terminology

Technical docs with exact identifiers benefit from strategy="hybrid" or hybrid=True.

Profile before overriding

Run a few queries with auto-selection first; only pin a strategy when benchmarks show a clear win.

Retrieval Configuration

Configure retrieval behaviour on agents

Vector Store

Pluggable embedding store with namespace support

Getting Started

Install

Core Concepts

Guides

Features

Models

Databases

Observability

Memory

Telemetry & Analytics

Knowledge

RAG

Persistence

Tools

Other Features

Developers

Configuration

Best Practices

Getting Started (No Code)

Retrieval Strategies

How It Works

Quick Start

Strategy Selection

Configuration Options

CLI Usage

Best Practices

Retrieval Configuration

Vector Store

​How It Works

​Quick Start

​Strategy Selection

​Configuration Options

​CLI Usage

​Best Practices

​Related

Retrieval Configuration

Vector Store

How It Works

Quick Start

Strategy Selection

Configuration Options

CLI Usage

Best Practices

Related