Reranker Providers

Fresh

Rerankers provide a second-pass processing step to improve search result precision. Disabled by default in Mem0 OSS.

Supported Providers

Provider	Description
Cohere	Production-grade reranker
Sentence Transformer	Open-source cross-encoder
Hugging Face	Custom reranker models
LLM Reranker	Use any LLM as reranker
Zero Entropy	Specialized reranker API

Configuration

from mem0 import Memory

config = {
    "reranker": {
        "provider": "cohere",
        "config": {
            "model": "rerank-english-v3.0",
            "api_key": "your-cohere-key",
        }
    }
}

memory = Memory.from_config(config)

Provider Examples

Cohere

config = {
    "reranker": {
        "provider": "cohere",
        "config": {
            "model": "rerank-english-v3.0",
        }
    }
}

Sentence Transformer

config = {
    "reranker": {
        "provider": "sentence_transformer",
        "config": {
            "model": "cross-encoder/ms-marco-MiniLM-L-6-v2",
        }
    }
}

LLM Reranker

config = {
    "reranker": {
        "provider": "llm_reranker",
        "config": {
            "model": "gpt-4.1-mini",
        }
    }
}

Zero Entropy

config = {
    "reranker": {
        "provider": "zero_entropy",
        "config": {
            "api_key": "your-zero-entropy-key",
        }
    }
}

Optimization Tips

• Limit top_k to 10-20 results to avoid latency
• Enable reranking when precision matters more than speed
• Use Cohere for production workloads
• Use Sentence Transformer for cost-free local reranking

Custom Prompts

You can customize the reranking prompt for domain-specific relevance scoring. See the Reranker Custom Prompts section for details.