vector-index-tuning

Optimizes HNSW parameters and quantization for target latency, recall, and memory.

1|1|Updated Apr 20, 2026

Search Tags:#recall #vector search #latency #memory optimization #quantization #hnsw

Ready to run

npx skills add https://github.com/svssdeva/agentic-skills --skill vector-index-tuning-svssdeva

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: vector-index-tuning
Source: https://github.com/svssdeva/agentic-skills/tree/main/ai/vector-index-tuning
Command: npx skills add https://github.com/svssdeva/agentic-skills --skill vector-index-tuning-svssdeva

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Vector search performance often degrades due to poorly chosen index structures and parameters, causing high latency, low recall, or excessive memory use in production systems.

Core Features & Use Cases

HNSW parameter optimization: Adjust M, efConstruction, and efSearch to reach a target recall/latency balance.
Quantization strategy selection: Choose FP16, INT8, product/binary quantization to reduce memory while preserving relevance.
Production scaling guidance: Apply different index approaches as datasets grow toward large-scale (e.g., billions of vectors).

Quick Start

Use the vector-index-tuning skill to recommend HNSW and quantization settings for my vector database to meet a target recall of 0.95 at under 10ms latency.

Frequently Asked Questions about vector-index-tuning

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I tune HNSW parameters to reduce vector search latency while maintaining high recall?▼

Tune HNSW parameters by adjusting M, efConstruction, and efSearch to balance graph connectivity and traversal depth, reducing vector search latency while maintaining high recall for your target dataset.

What is the best quantization approach to optimize vector memory consumption?▼

The best quantization approach to optimize vector memory consumption depends on your constraints: choose FP16, INT8, product, or binary quantization to compress vectors while preserving search relevance.

How do I scale vector search infrastructure for very large datasets with billions of vectors?▼

Scale vector search infrastructure for billions of vectors by applying different index approaches and quantization strategies to manage memory consumption and latency as dataset size grows.

How do I choose the right vector index type for my specific latency and recall requirements?▼

Choose the right vector index type by evaluating your specific latency, recall, and memory constraints, then selecting HNSW configurations and quantization levels that align with those production targets.

Why does my vector search have low recall even after configuring the index?▼

Vector search recall drops when HNSW parameters like M and efSearch are set too low for your dataset complexity; increasing these values expands the search radius to capture more relevant results.