Skip to content

Introduction

TensorTensor builds high-quality HNSW indexes from your vector data on NVIDIA GPUs and returns a ready-to-load file. You submit a job, we build the index in batch on a GPU, and we email you a download link when it’s ready.

In-process search is already fast. The only hard part was building the index. We solved that.

Your data, your server, your index.

  1. Host your vectors at a publicly reachable HTTPS URL (raw float32).
  2. Submit a build job with that URL and your dimensions (768, 1024, or 1536).
  3. Receive a jobId. Builds run in batch within a 24-hour window.
  4. When the build completes, you get an email with a download link to the HNSW file.
  5. Load the file with the standard HNSW library in your language and run searches.