# Unum

> Unum builds open-source search, storage and communication engines, USearch, UStore and UCall, on its own kernel libraries. Inside ClickHouse, DuckDB and 100M+ devices.

Every page is also served as markdown at its URL plus `.md`, or at its URL itself with an `Accept: text/markdown` header; add `?q=` with search terms to get only the sections containing all of them, and each engine or library has its own index at `/docs/<name>/llms.txt` and all of its pages joined in `/docs/<name>/llms-full.txt`.

The same search and pages are tools of the MCP server at https://www.unum.cloud/mcp over Streamable HTTP: `search_docs`, `read_page` and `list_docs`.

## Engines

- [USearch | One Index for Every Modality](https://www.unum.cloud/engines/usearch.md): USearch, Unum’s open-source search engine inside ClickHouse and DuckDB: one proximity graph for search, attention and sparse algebra over texts, genomes, maps and embeddings
- [UStore | One Store for Transactions and Analytics](https://www.unum.cloud/engines/ustore.md): UStore, Unum’s embedded database for blobs, documents, graphs, strings and pages: one transaction across every modality, scans and random sampling over the same copy, on local NVMe and GPUs
- [UForm | One Model for Every Modality](https://www.unum.cloud/engines/uform.md): UForm, Unum’s tokenizer-free multimodal encoders trained from scratch in CUDA: raw bytes of text, images, video and audio into one embedding

## USearch

Search, attention and sparse algebra behind one C ABI, in RAM, VRAM or on NVMe. Source: https://github.com/unum-cloud/USearch

- [Everything](https://www.unum.cloud/docs/usearch/llms-full.txt): all of the pages below in one file
- [Overview](https://www.unum.cloud/docs/usearch.md)
- [C](https://www.unum.cloud/docs/usearch/c.md)
- [Python](https://www.unum.cloud/docs/usearch/python.md)
- [Rust](https://www.unum.cloud/docs/usearch/rust.md)
- [JavaScript](https://www.unum.cloud/docs/usearch/javascript.md)
- [Go](https://www.unum.cloud/docs/usearch/go.md)
- [Swift](https://www.unum.cloud/docs/usearch/swift.md)
- [Java](https://www.unum.cloud/docs/usearch/java.md)
- [C#](https://www.unum.cloud/docs/usearch/csharp.md)
- [Zig](https://www.unum.cloud/docs/usearch/zig.md)
- [Wolfram](https://www.unum.cloud/docs/usearch/wolfram.md)
- [Mojo](https://www.unum.cloud/docs/usearch/mojo.md)
- [Contributing](https://www.unum.cloud/docs/usearch/contributing.md): hooks, code style and naming
- [C++](https://www.unum.cloud/docs/usearch/cpp.md): every topic, with its symbols

## UStore

Transactional blobs, documents and graphs on local NVMe, behind one C ABI. Source: https://github.com/unum-cloud/UStore

- [Everything](https://www.unum.cloud/docs/ustore/llms-full.txt): all of the pages below in one file
- [Overview](https://www.unum.cloud/docs/ustore.md)
- [C](https://www.unum.cloud/docs/ustore/c.md)
- [C++](https://www.unum.cloud/docs/ustore/cpp.md)
- [Python](https://www.unum.cloud/docs/ustore/python.md)
- [Rust](https://www.unum.cloud/docs/ustore/rust.md)
- [Mojo](https://www.unum.cloud/docs/ustore/mojo.md)
- [Contributing](https://www.unum.cloud/docs/ustore/contributing.md): hooks, code style and naming

## UForm

Distributed training and inference pipeline for multi-modal perception. Source: https://github.com/unum-cloud/UForm

- [Everything](https://www.unum.cloud/docs/uform/llms-full.txt): all of the pages below in one file
- [Overview](https://www.unum.cloud/docs/uform.md)
- [C](https://www.unum.cloud/docs/uform/c.md)
- [Python](https://www.unum.cloud/docs/uform/python.md)
- [Rust](https://www.unum.cloud/docs/uform/rust.md)
- [JavaScript](https://www.unum.cloud/docs/uform/javascript.md)
- [Swift](https://www.unum.cloud/docs/uform/swift.md)
- [Mojo](https://www.unum.cloud/docs/uform/mojo.md)
- [Contributing](https://www.unum.cloud/docs/uform/contributing.md): hooks, code style and naming

## UCall

Client-server communication engine built around Noise, io_uring and SIMD. Source: https://github.com/unum-cloud/UCall

- [Everything](https://www.unum.cloud/docs/ucall/llms-full.txt): all of the pages below in one file
- [Overview](https://www.unum.cloud/docs/ucall.md)
- [C](https://www.unum.cloud/docs/ucall/c.md)
- [Python](https://www.unum.cloud/docs/ucall/python.md)
- [Rust](https://www.unum.cloud/docs/ucall/rust.md)
- [Mojo](https://www.unum.cloud/docs/ucall/mojo.md)
- [Contributing](https://www.unum.cloud/docs/ucall/contributing.md): hooks, code style and naming

## NumKong

Mixed-precision math for every CPU and GPU. Source: https://github.com/ashvardanian/NumKong

- [Everything](https://www.unum.cloud/docs/numkong/llms-full.txt): all of the pages below in one file
- [Overview](https://www.unum.cloud/docs/numkong.md)
- [C](https://www.unum.cloud/docs/numkong/c.md)
- [C++](https://www.unum.cloud/docs/numkong/cpp.md)
- [Python](https://www.unum.cloud/docs/numkong/python.md)
- [Rust](https://www.unum.cloud/docs/numkong/rust.md)
- [JavaScript](https://www.unum.cloud/docs/numkong/javascript.md)
- [Go](https://www.unum.cloud/docs/numkong/go.md)
- [Swift](https://www.unum.cloud/docs/numkong/swift.md)
- [Contributing](https://www.unum.cloud/docs/numkong/contributing.md): hooks, code style and naming

## StringZilla

SIMD and GPU-accelerated strings. Source: https://github.com/ashvardanian/StringZilla

- [Everything](https://www.unum.cloud/docs/stringzilla/llms-full.txt): all of the pages below in one file
- [Overview](https://www.unum.cloud/docs/stringzilla.md)
- [C](https://www.unum.cloud/docs/stringzilla/c.md)
- [C++](https://www.unum.cloud/docs/stringzilla/cpp.md)
- [Python](https://www.unum.cloud/docs/stringzilla/python.md)
- [Rust](https://www.unum.cloud/docs/stringzilla/rust.md)
- [JavaScript](https://www.unum.cloud/docs/stringzilla/javascript.md)
- [Go](https://www.unum.cloud/docs/stringzilla/go.md)
- [Swift](https://www.unum.cloud/docs/stringzilla/swift.md)
- [Java](https://www.unum.cloud/docs/stringzilla/java.md)
- [C#](https://www.unum.cloud/docs/stringzilla/csharp.md)
- [Contributing](https://www.unum.cloud/docs/stringzilla/contributing.md): hooks, code style and naming

## ForkUnion

NUMA-aware fork-join parallelism. Source: https://github.com/ashvardanian/ForkUnion

- [Everything](https://www.unum.cloud/docs/forkunion/llms-full.txt): all of the pages below in one file
- [Overview](https://www.unum.cloud/docs/forkunion.md)
- [C++](https://www.unum.cloud/docs/forkunion/cpp.md)
- [C](https://www.unum.cloud/docs/forkunion/c.md)
- [Rust](https://www.unum.cloud/docs/forkunion/rust.md)
- [Zig](https://www.unum.cloud/docs/forkunion/zig.md)
- [Mojo](https://www.unum.cloud/docs/forkunion/mojo.md)
- [Contributing](https://www.unum.cloud/docs/forkunion/contributing.md): hooks, code style and naming

## SmashTable

Transactions across in-memory containers. Source: https://github.com/ashvardanian/SmashTable

- [Everything](https://www.unum.cloud/docs/smashtable/llms-full.txt): all of the pages below in one file
- [Overview](https://www.unum.cloud/docs/smashtable.md)
- [C++](https://www.unum.cloud/docs/smashtable/cpp.md)
- [Python](https://www.unum.cloud/docs/smashtable/python.md)
- [Contributing](https://www.unum.cloud/docs/smashtable/contributing.md): hooks, code style and naming

## Optional

- [Blog feed](https://www.unum.cloud/blog/rss.xml): every post as RSS
- [Home](https://www.unum.cloud/index.md): six assumptions about hardware and what changes without them
- [Docs](https://www.unum.cloud/docs.md): five engineering positions and the mechanisms behind them
- [About](https://www.unum.cloud/about.md): the company and its history
- [Notes](https://www.unum.cloud/notes.md): open science and open datasets
- [Networks Processing Performance Report](https://www.unum.cloud/blog/graphs.md): Benchmarking graph traversal performance on persistent storage against NetworkX, evaluating random access patterns on Patent Citations, Gene Regulatory Networks, and Human Brain connectomes.
- [Why Build a Custom DBMS?](https://www.unum.cloud/blog/why-build-dbms.md): Analyzing the widening gap between modern NVMe SSD throughput and legacy database software performance, and why we decided to build a new storage engine from scratch.
- [Text Indexing in ElasticSearch, MongoDB and UDisk](https://www.unum.cloud/blog/texts.md): Comparing indexing performance of ElasticSearch, MongoDB, and UDisk on a 20GB text dataset, revealing the massive overhead of Java-based search engines.
- [Our Recipe for Fast Computational Storage](https://www.unum.cloud/blog/bottlenecks.md): Exploring the limits of modern hardware through Assembly optimizations, CUDA kernels, and algorithmic improvements to eliminate IO and compute bottlenecks in database engines.
- [Yahoo Cloud Serving Benchmark for Key-Value Stores](https://www.unum.cloud/blog/ycsb-keyval.md): A deep dive into the Yahoo Cloud Serving Benchmark to evaluate NoSQL database performance, identifying the strengths and weaknesses of RocksDB, LevelDB, and WiredTiger.
- [UDisk Crushing Benchmarks: YCSB, pt. 2](https://www.unum.cloud/blog/ycsb-udisk.md): Benchmarking UDisk against RocksDB, LevelDB, and WiredTiger on 1TB datasets, demonstrating superior performance in read-heavy and write-heavy workloads.
- [2021 in Database Startups: Gold Rush!](https://www.unum.cloud/blog/dbms-startups.md): A retrospective on the database industry in 2021, analyzing the technical architecture, funding rounds, and market promises of the hottest new database startups.
- [CPU vs GPU Benchmarks Series](https://www.unum.cloud/blog/cpu-vs-gpu.md): Comparing the number-crunching capabilities of AMD Threadripper PRO 3995WX and Nvidia RTX 3090 across matrix multiplication, convolution, and other compute-intensive tasks.
- [50x Faster NumPy on GPUs with CuPy](https://www.unum.cloud/blog/cupy.md): Accelerating NumPy workflows by 50x using CuPy on Nvidia RTX 3090, benchmarking BLAS operations against CPU-based implementations on AMD Threadripper.
- [Unbranding and Extending the Yahoo Cloud Serving Benchmark](https://www.unum.cloud/blog/ucsb.md): Introducing the Unum Cloud Serving Benchmark (UCSB), a modernized and extended version of YCSB designed to rigorously evaluate the performance of next-generation storage engines.
- [Beating RocksDB by up to 7x in almost every workload](https://www.unum.cloud/blog/ucsb-10tb.md): Pushing the limits of UDisk and RocksDB on massive 10TB datasets with 64 threads and 1TB of RAM, achieving up to 7x performance gains in almost every workload.
- [Billion Rows in Pandas? CuDF, Modin, Arrow, Spark](https://www.unum.cloud/blog/pandas.md): Evaluating in-memory tabular processing frameworks including CuDF, Modin, Arrow, and Spark on billion-row taxi ride datasets to find the fastest Pandas alternative.
- [Beating OpenAI CLIP with 100x less data and compute](https://www.unum.cloud/blog/efficient-multimodality.md): Demonstrating how to train state-of-the-art Vision-Language transformers for semantic search using 100x less data and compute than OpenAI CLIP.
- [Making the Edge Efficient with Gcore and Intel](https://www.unum.cloud/blog/intel-gcore-unum.md): Partnering with Gcore and Intel to deploy the world most efficient data-processing infrastructure at the edge, powered by Unum assembly-optimized engines.
- [UForm v2: now in 21 languages](https://www.unum.cloud/blog/uform-graphcore.md): Releasing UForm v2 with support for 21 languages, delivering state-of-the-art efficiency in AI-based text-to-image search and multimodal embeddings.
- [USearch: 10x - 100x Faster than Meta's FAISS](https://www.unum.cloud/blog/scaling-vector-search-with-intel.md): Scaling Retrieval Augmented Generation with the USearch vector search engine on Intel Sapphire Rapids CPUs on Gcore, outperforming Meta's FAISS library by 10x - 100x on a scale of 100M to 1B vectors.
- [FP8 Search & KV-Caching in USearch](https://www.unum.cloud/blog/float8.md): GPUs moved to FP8. CPUs didn't. 200'000 lines of SIMD kernels later, USearch bridges both worlds — native Float8 Vector Search and KV Caching for the upcoming Arm-based NVIDIA Vera racks, Intel Xeons, open-source RISC-V chips, IBM Mainframes, Chinese LoongArch, and even WebAssembly in the browser — for higher throughput at 99% recall and half the memory cost
- [Changing Our Development Practices with Time](https://www.unum.cloud/blog/development-practices.md): A decade in, AI agents made writing code cheap and keeping it coherent expensive. How that changed our Git branches and releases, moved eight repositories into lockstep, turned pre-commit hooks into design guardrails, made model checking routine, and what we still have not figured out