Crate elid 

ELID - Embedding Locality IDentifier

ELID enables vector search without a vector store by encoding high-dimensional embeddings into sortable string IDs that preserve locality. Similar vectors produce similar IDs, allowing you to use standard database indexes for similarity search.

ELID also includes a complete suite of fast, zero-dependency string similarity algorithms.

Feature Sets

Embedding Encoding (embeddings feature)

Convert embeddings from any ML model into compact, sortable identifiers:

• Mini128: 128-bit SimHash using signed random projections (fast, Hamming distance)
• Morton10x10: Z-order curve encoding (database range queries)
• Hilbert10x10: Hilbert curve encoding (maximum locality preservation)

String Similarity (strings feature, default)

• Levenshtein Distance: Classic edit distance algorithm
• Normalized Levenshtein: Returns similarity as a value between 0.0 and 1.0
• Jaro-Winkler Similarity: Better for short strings like names
• Hamming Distance: For equal-length strings
• Optimal String Alignment (OSA): Levenshtein with transpositions
• SimHash: Locality-sensitive hashing for string similarity queries

Feature Flags

• strings (default): Zero-dependency string similarity algorithms
• embeddings (default): Vector encoding with Mini128, Morton, and Hilbert profiles
• models: Base ONNX model support using tract-onnx (WASM compatible)
• models-text: Text embedding models (Model2Vec potion-base-8M)
• models-image: Image embedding models (MobileNetV3-Small)
• wasm: WebAssembly bindings (includes embeddings)
• python: Python bindings via PyO3 (includes embeddings + numpy)
• ffi: C FFI bindings

Embedding Encoding Example

use elid::embeddings::{encode, Profile, hamming_distance};

// Get embeddings from your ML model
let embedding1 = model.embed("Hello, world!")?;
let embedding2 = model.embed("Hello, universe!")?;

// Encode to sortable ELIDs
let profile = Profile::default(); // Mini128
let elid1 = encode(&embedding1, &profile)?;
let elid2 = encode(&embedding2, &profile)?;

// Compare via Hamming distance (lower = more similar)
let distance = hamming_distance(&elid1, &elid2)?;

String Similarity Example

use elid::{levenshtein, normalized_levenshtein, jaro_winkler, simhash, simhash_similarity};

let distance = levenshtein("kitten", "sitting");
assert_eq!(distance, 3);

let similarity = normalized_levenshtein("kitten", "sitting");
assert!(similarity > 0.5 && similarity < 0.7);

let jw_similarity = jaro_winkler("martha", "marhta");
assert!(jw_similarity > 0.9);

// SimHash for numeric database queries
let hash1 = simhash("iPhone 14");
let hash2 = simhash("iPhone 15");
let sim = simhash_similarity("iPhone 14", "iPhone 15");
assert!(sim > 0.8);

Modules

embeddings

Embedding encoding module for ELID

Structs

SimilarityOpts

Options for configuring string similarity algorithms

Functions

best_match

Compute the best matching similarity between two strings using multiple algorithms and return the highest score.

find_best_match

Find the best match for a query string in a list of candidates.

find_matches_above_threshold

Find all matches above a threshold score.

find_similar_hashes

Find all items within a given SimHash distance threshold.

hamming

Compute the Hamming distance between two strings.

jaro

Compute the Jaro similarity between two strings.

jaro_winkler

Compute the Jaro-Winkler similarity between two strings.

jaro_winkler_with_prefix

Compute the Jaro-Winkler similarity with a custom prefix scale.

levenshtein

Compute the Levenshtein distance between two strings.

levenshtein_with_opts

Compute Levenshtein distance with configurable options.

normalized_hamming

Compute the normalized Hamming similarity between two strings.

normalized_levenshtein

Compute the normalized Levenshtein similarity between two strings.

normalized_osa

Compute the normalized OSA similarity between two strings.

osa_distance

Compute the Optimal String Alignment distance between two strings.

simhash

Compute the SimHash fingerprint of a string.

simhash_distance

Compute the Hamming distance between two SimHash values.

simhash_similarity

Compute the normalized SimHash similarity between two strings.