100 Embeddings & Vector Search resources for developers

Recommended starting point for general-purpose embeddings. Offers a 1536-dimension vector with a 'dimensions' parameter to reduce size without retraining.

High-performance model optimized for retrieval tasks. Consistently ranks at the top of the MTEB leaderboard for document search.

Features a 'compression' parameter for binary or int8 quantization, significantly reducing storage costs in vector databases.

Standard Python library for running open-source models like BGE-large or All-MiniLM-L6-v2 locally or on private infrastructure.

Supports an 8k token context window, making it suitable for long-form document embedding where chunking might lose context.

Massive Text Embedding Benchmark. Use this to compare model performance across specific tasks like clustering, retrieval, and summarization.

Convert Hugging Face models to ONNX format to run embeddings in Node.js or C# environments with minimal latency.

A versatile model supporting multi-lingual, multi-functional (dense/sparse), and multi-granularity retrieval within a single embedding.

A high-throughput, MIT-licensed inference server for deploying open-source embedding models via a REST API.

Enterprise-grade embedding models integrated with GCP, offering high rate limits and managed infrastructure.

Extension enabling vector similarity search in Postgres. Best for teams already using RDS or Supabase to avoid adding new infra.

Rust-based vector database. Offers advanced payload filtering and high-performance HNSW indexing for production scale.

Managed vector database that scales based on usage. Ideal for applications with unpredictable traffic and large datasets.

Open-source vector DB with a GraphQL interface and native support for multi-modal data (images, video, text).

Lightweight, developer-centric database often used for local RAG development and small-scale deployments.

Cloud-native vector database designed for billion-scale vector search with decoupled storage and compute.

Serverless, disk-based vector database that stores data in the Lance format, optimized for random access and large-scale AI data.

Library for efficient similarity search. Essential for building custom indexing pipelines or running in-memory searches.

Utilizes k-NN plugin for vector search. Best for organizations already invested in the ELK stack for logging and search.

The Vector Library for Redis, enabling low-latency vector indexing and search directly within a Redis instance.

Combines traditional keyword search with semantic vector search using Reciprocal Rank Fusion (RRF) for better accuracy.

A cross-encoder model used as a second stage to re-score the top results from a vector search for higher precision.

The industry-standard algorithm for approximate nearest neighbor (ANN) search, balancing speed and recall.

A compression technique that divides vectors into sub-vectors to reduce memory usage by up to 90% at a slight cost to accuracy.

A retrieval strategy that re-ranks results to reduce redundancy and increase the diversity of the returned items.

Technique where you embed small chunks for search but return the full parent document to the LLM for better context.

Using an LLM to generate multiple versions of a user's query to catch more relevant vectors during retrieval.

Applying hard filters (e.g., date > 2023) before performing the vector similarity search to narrow the search space.

Filtering out irrelevant parts of retrieved documents before passing them to an LLM to save on token costs.

Assigning multiple vectors to a single document (e.g., summary vector and full text vector) to improve retrieval hits.