{"slug": "usearch", "name": "USearch", "category": "search", "category_name": "Search", "description": "", "recommenders": 5, "commenters": 6, "warners": 0, "score": 4.394, "rising": -1.64, "first": "2026-01-14", "latest": "2026-03-06", "why": "It's plenty fast and is happy working with vectors on disk", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 5, "neutral": 0, "compare": 1, "warn": 0}, "spark": [0, 0, 2, 2, 1, 0, 0, 0, 0, 0, 0, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "It's plenty fast and is happy working with vectors on disk", "kind": "", "icon": true, "domain": null, "github_repo": "github.com/unum-cloud/usearch", "official_url": "https://github.com/unum-cloud/usearch", "pricing": "", "aliases": ["USearch"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 756559920, "item_id": 47284995, "url": "https://news.ycombinator.com/item?id=47284995", "by": "sdenton4", "date": "2026-03-06", "stance": "recommend", "quote": "It's plenty fast and is happy working with vectors on disk", "excerpt": "Depending on how 'one-off' the query is, sequential read is the right answer. The alternative is indexing the data for ANN, which will generally require doing the equivalent of many queries across the dataset. On the bright side, smart folks have already thought pretty hard about this. In my work, I ended up picking usearch for large-scale vector storage and ANN search. It's plenty fast and is happy working with vect", "context": "Uses it at work", "use_case": "Vector storage and ANN search", "switched_from": "", "story": "Querying 3B Vectors"}, {"id": 752788176, "item_id": 47049261, "url": "https://news.ycombinator.com/item?id=47049261", "by": "stevesimmons", "date": "2026-02-17", "stance": "recommend", "quote": "It gets similar performance and is very convenient.", "excerpt": "USearch has a sqlite extension that supports various metrics on including Hamming distance on standard sqlite BLOB columns. It gets similar performance and is very convenient. (There's also an indexed variant that does faster lookups, but it uses a special virtual table layout that constrains the types of the other columns in the table.) See https://github.com/unum-cloud/USearch. pip-installable for Python users.", "context": "", "use_case": "vector search in SQLite", "switched_from": "", "story": "Hamming Distance for Hybrid Search in SQLite"}, {"id": 752381040, "item_id": 47023815, "url": "https://news.ycombinator.com/item?id=47023815", "by": "luoxiaojian", "date": "2026-02-15", "stance": "recommend", "quote": "USearch has some impressive work", "excerpt": "Author here. Appreciate the context\u2014just wanted to add some perspective on the 8K QPS figure: in the VectorDBBench setting we used (10M, 768d, on comparable hardware to the previous leader), we're seeing double their throughput\u2014so it's far from trivial on that playing field. That said, self-reported numbers only go so far\u2014it'd be great to see USearch in more third-party benchmarks like VectorDBBench or ANN-Benchmarks", "context": "", "use_case": "Vector search library", "switched_from": "", "story": "zvec: embedded vector database"}, {"id": 746062096, "item_id": 46628881, "url": "https://news.ycombinator.com/item?id=46628881", "by": "ehsanu1", "date": "2026-01-14", "stance": "recommend", "quote": "Embedded usearch vector database.", "excerpt": "Embedded usearch vector database. https://github.com/unum-cloud/USearch", "context": "", "use_case": "local RAG vector search", "switched_from": "", "story": "Ask HN: How are you doing RAG locally?"}, {"id": 745927280, "item_id": 46620455, "url": "https://news.ycombinator.com/item?id=46620455", "by": "mips_avatar", "date": "2026-01-14", "stance": "recommend", "quote": "Mostly usearch since pgvector had some perf problems", "excerpt": "Mostly usearch since pgvector had some perf problems, though I use postgres during my ingestion stage since postgis is so good. I segment my hnsw indexes by multipe h3 levels.", "context": "", "use_case": "vector index for geocoder", "switched_from": "pgvector", "story": "I built a geocoder for AI agents because I couldn't afford Google Maps"}]}], "caveats": [], "who": [{"by": "sdenton4", "context": "Uses it at work", "url": "https://news.ycombinator.com/item?id=47284995", "date": "2026-03-06"}], "tags": [], "use_cases": ["vector storage and ann search", "vector search in sqlite", "vector search library", "local rag vector search", "vector index for geocoder"], "compared": [{"slug": "pgvector", "name": "pgvector", "co": 1, "switched_from": 1, "switched_to": 0}], "makers": [{"id": 752320112, "item_id": 47020007, "url": "https://news.ycombinator.com/item?id=47020007", "by": "ashvardanian", "date": "2026-02-14", "stance": "recommend", "quote": "USearch used ~100 custom kernels for different data types", "excerpt": "8K QPS is probably quite trivial on their setup and a 10M dataset. I rarely use comparably small instances & datasets in my benchmarks, but on 100M-1B datasets on a larger dual-socket server, 100K QPS was easily achievable in 2023: https://www.unum.cloud/blog/2023-11-07-scaling-vector-search-with-intel ;) Typically, the recipe is to keep the hot parts of the data structure in SRAM in CPU caches and a lot of SIMD. At ", "context": "", "use_case": "Vector similarity search", "switched_from": "", "story": "zvec: embedded vector database"}], "threads": [{"url": "https://news.ycombinator.com/item?id=47000535", "title": "zvec: embedded vector database", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=47231871", "title": "Querying 3B Vectors", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47021711", "title": "Hamming Distance for Hybrid Search in SQLite", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46616529", "title": "Ask HN: How are you doing RAG locally?", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46619947", "title": "I built a geocoder for AI agents because I couldn't afford Google Maps", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {"recommend": 2}, "2026-02": {"compare": 1, "recommend": 2}, "2026-03": {"recommend": 1}, "2026-04": {}, "2026-05": {}, "2026-06": {}, "2026-07": {}, "2026-08": {}, "2026-09": {}, "2026-10": {}}, "quotes": [{"id": 756559920, "item_id": 47284995, "url": "https://news.ycombinator.com/item?id=47284995", "by": "sdenton4", "date": "2026-03-06", "stance": "recommend", "quote": "It's plenty fast and is happy working with vectors on disk", "excerpt": "Depending on how 'one-off' the query is, sequential read is the right answer. The alternative is indexing the data for ANN, which will generally require doing the equivalent of many queries across the dataset. On the bright side, smart folks have already thought pretty hard about this. In my work, I ended up picking usearch for large-scale vector storage and ANN search. It's plenty fast and is happy working with vect", "context": "Uses it at work", "use_case": "Vector storage and ANN search", "switched_from": "", "story": "Querying 3B Vectors"}, {"id": 752788176, "item_id": 47049261, "url": "https://news.ycombinator.com/item?id=47049261", "by": "stevesimmons", "date": "2026-02-17", "stance": "recommend", "quote": "It gets similar performance and is very convenient.", "excerpt": "USearch has a sqlite extension that supports various metrics on including Hamming distance on standard sqlite BLOB columns. It gets similar performance and is very convenient. (There's also an indexed variant that does faster lookups, but it uses a special virtual table layout that constrains the types of the other columns in the table.) See https://github.com/unum-cloud/USearch. pip-installable for Python users.", "context": "", "use_case": "vector search in SQLite", "switched_from": "", "story": "Hamming Distance for Hybrid Search in SQLite"}, {"id": 752381040, "item_id": 47023815, "url": "https://news.ycombinator.com/item?id=47023815", "by": "luoxiaojian", "date": "2026-02-15", "stance": "recommend", "quote": "USearch has some impressive work", "excerpt": "Author here. Appreciate the context\u2014just wanted to add some perspective on the 8K QPS figure: in the VectorDBBench setting we used (10M, 768d, on comparable hardware to the previous leader), we're seeing double their throughput\u2014so it's far from trivial on that playing field. That said, self-reported numbers only go so far\u2014it'd be great to see USearch in more third-party benchmarks like VectorDBBench or ANN-Benchmarks", "context": "", "use_case": "Vector search library", "switched_from": "", "story": "zvec: embedded vector database"}, {"id": 752302609, "item_id": 47018913, "url": "https://news.ycombinator.com/item?id=47018913", "by": "clemlesne", "date": "2026-02-14", "stance": "compare", "quote": "Did someone compared with uSearch", "excerpt": "Did someone compared with uSearch (https://github.com/unum-cloud/USearch)?", "context": "", "use_case": "", "switched_from": "", "story": "zvec: embedded vector database"}, {"id": 746062096, "item_id": 46628881, "url": "https://news.ycombinator.com/item?id=46628881", "by": "ehsanu1", "date": "2026-01-14", "stance": "recommend", "quote": "Embedded usearch vector database.", "excerpt": "Embedded usearch vector database. https://github.com/unum-cloud/USearch", "context": "", "use_case": "local RAG vector search", "switched_from": "", "story": "Ask HN: How are you doing RAG locally?"}, {"id": 745927280, "item_id": 46620455, "url": "https://news.ycombinator.com/item?id=46620455", "by": "mips_avatar", "date": "2026-01-14", "stance": "recommend", "quote": "Mostly usearch since pgvector had some perf problems", "excerpt": "Mostly usearch since pgvector had some perf problems, though I use postgres during my ingestion stage since postgis is so good. I segment my hnsw indexes by multipe h3 levels.", "context": "", "use_case": "vector index for geocoder", "switched_from": "pgvector", "story": "I built a geocoder for AI agents because I couldn't afford Google Maps"}]}