Praison AI

Retrieval Methods

Configure retrieval strategies for knowledge bases

Tune top-k, similarity thresholds, and reranking so retrieval quality matches your use case.

python
from praisonaiagents import Agent, KnowledgeConfig

agent = Agent(
    name="Researcher",
    instructions="Cite retrieved passages in every answer.",
    knowledge=KnowledgeConfig(sources=["docs/"], top_k=5, similarity_threshold=0.7),
)
agent.start("Summarise the security section of our handbook.")

The user submits a query; the agent pulls the best-matching passages before composing a reply.

Diagram

Retrieval

📋 Query

🔍 Rank passages

✅ Top matches

Quick Start

  1. Simple Usage

    Retrieve the top matching passages with a similarity floor.

    python
    from praisonaiagents import Knowledge

knowledge = Knowledge(sources=["docs/"], top_k=5, similarity_threshold=0.7)
  2. With Configuration

    Switch to hybrid search to blend keyword and semantic scoring.

    python
    from praisonaiagents import Knowledge

knowledge = Knowledge(
    sources=["docs/"],
    retrieval_strategy="hybrid",
    keyword_weight=0.3,
    semantic_weight=0.7,
)

How It Works

Diagram
KnowledgeAgentUserKnowledgeAgentUserSubmit queryRank passages by strategyTop-k passagesAnswer citing the passages

Basic Retrieval

python
from praisonaiagents import Knowledge

knowledge = Knowledge(
    sources=["docs/"],
    top_k=5,                    # Number of results
    similarity_threshold=0.7    # Minimum similarity
)

Retrieval Strategies

Similarity Search

python
knowledge = Knowledge(
    sources=["docs/"],
    retrieval_strategy="similarity",
    top_k=5
)

MMR (Maximal Marginal Relevance)

python
knowledge = Knowledge(
    sources=["docs/"],
    retrieval_strategy="mmr",
    top_k=5,
    diversity=0.3  # Balance relevance vs diversity
)

Hybrid Search

python
knowledge = Knowledge(
    sources=["docs/"],
    retrieval_strategy="hybrid",
    keyword_weight=0.3,
    semantic_weight=0.7
)

Query Enhancement

Use QueryRewriterAgent for better retrieval:

python
from praisonaiagents import Agent, Knowledge, QueryRewriterAgent

# Rewrite queries for better retrieval
rewriter = QueryRewriterAgent()

knowledge = Knowledge(sources=["docs/"])

agent = Agent(
    name="Assistant",
    knowledge=knowledge,
    query_rewriter=rewriter
)

Best Practices

Pick the strategy that matches the query type

Use similarity for semantic questions, hybrid when exact keywords matter, and mmr when you need diverse rather than near-duplicate passages.

Balance diversity with MMR

diversity=0.3 trades a little relevance for broader coverage. Raise it when top results repeat the same point.

Rewrite vague queries before retrieval

QueryRewriterAgent reshapes short or ambiguous questions into retrieval-friendly form, improving recall without changing the user’s intent.

Tune top_k and threshold together

A high top_k with a low similarity_threshold floods the context; a low top_k with a high threshold can starve it. Adjust both when tuning quality.

Related

Knowledge Module

Full API reference

 QueryRewriterAgent

Query optimisation