MRHI Vector Search

Try the fastest vector search against your own data without running your own infrastructure. We are looking for a company to purchase the full IP from us. We are the first truly real-time vector database!

Quick Start REST API

Headline Metrics

150,000+ CPU QPS

Up to 150,000+ QPS on the public 10k / 128d CPU benchmark with rust (shared).

1M+ CUDA QPS

Up to 1m+ QPS on CUDA in the published benchmark results.

8.6M+ Inserts/Sec

Up to 8.6M+ inserts/sec on the public 10k / 128d benchmark with rust (shared).

99.94% - 100% Recall

Published public and synthetic benchmark tables currently span 99.94% to 100.00% recall.

562x Faster Builds

Up to 562x faster build throughput than local Qdrant on the public 100k CPU comparison.

Benchmarked to 100M

Benchmarked up to 100,000,000 vectors in the published large-scale runs.

What You Get

JavaScript SDK

Import the browser-ready SDK from /api/demo/sdk.js and start indexing in a few lines.

Anonymous Demo Tokens

Create a token, save it, and reuse the same session later from your app or scripts.

Filtering + Batch Search

Run exact-match metadata filters and use both search() and searchMany().

Live Limits Endpoint

Read the current public limits from /api/demo/limits instead of hard-coding them.

Performance Metrics

Every response includes request timing plus operation-specific timings like searchMs or addManyMs.

Demo Basics

Base URL: https://mrhisearch.com/api/demo
SDK URL: https://mrhisearch.com/api/demo/sdk.js
OpenAPI: https://mrhisearch.com/api/openapi
Live limits: https://mrhisearch.com/api/demo/limits

Every demo session is tied to a token. That token owns one temporary index. You can add up to the current public limit for that token, reuse the token later, clear the vectors while keeping the index configuration, or fully reset the index to switch dimensions.

<script type="module">
  import { createMRHIDemoClient } from "https://mrhisearch.com/api/demo/sdk.js";

  const client = createMRHIDemoClient({
    baseUrl: "https://mrhisearch.com/api/demo",
    dimensions: 3,
  });
  const token = await client.getToken();

  console.log("Save this token:", token);

  await client.addMany([
    { id: "doc-1", vector: [1, 0, 0], metadata: { category: "news" } },
    { id: "doc-2", vector: [0, 1, 0], metadata: { category: "sports" } },
  ]);

  const search = await client.search([1, 0, 0], {
    topK: 2,
    includeMetadata: true,
    filter: { category: "news" },
  });

  console.log(search.results);
  console.log(search.metrics);
</script>

Use the guides in the sidebar for the full SDK flow, raw REST examples, and the exact endpoint surface.

Benchmark Snapshot

These charts were generated from a self-run benchmark using a third-party, open-source benchmark suite based on the ann-benchmarks repository.

QPS at 10k vectors

QPS benchmark at 10k vectors

Build speed across the full run

Build speed benchmark across the full run

For the full benchmark chart set, see Performance.

Full IP Availability

We are actively looking to sell the full MRHI IP, which includes the complete system and full test suite.

MRHI is compatible with CPUs
MRHI is compatible with CUDA GPUs
MRHI is compatible with Apple Metal
MRHI is heavily optimized for performance
We are also open to licensing

For acquisition or licensing discussions, email

acquisitions@mrhisearch.com.

More detail: Licensing & Acquisition.