Skip to content
Rush Labs
All work

PixaFind

Offline AI search that finds documents by meaning and photos by description. We worked on the search pipeline, GPU inference and architecture with the PixaLabs team.

Client
PixaLabs
Focus
Retrieval, On-device inference
Our part
Search pipeline, GPU inference, Architecture
File formats searched by contentFrom PixaFind’s product page
76+
Typical search time once indexedFrom PixaFind’s product page
<100 ms
Files sent to the cloudEvery model runs on the user’s machine
0

The problem

People forget what a file was called, not what it was about. PixaLabs wanted search that understands meaning and finds photos from a plain description, on an ordinary Windows or Mac computer, without sending a single file to a server.

What we worked on

  • One index for text and images

    Local embedding models turn documents and photos into vectors in a shared space, so one plain-English query can match a paragraph inside a PDF or a picture with no tags.

  • GPU when it is there, CPU when it is not

    Inference uses an NVIDIA GPU to speed up indexing and falls back to the CPU on machines without one, so the same app works on a gaming PC and a work laptop.

  • Indexing that stays out of the way

    Indexing runs in the background at low priority. Files that are already indexed are searchable while the rest finish, and the computer stays responsive.

  • Private by architecture

    The models, the index and the search all live on the user’s machine. Privacy is a property of the design, not a policy promise.

Outcome

PixaFind ships for Windows with a macOS beta, on free and paid plans. It searches inside more than 76 file formats and answers typical queries in under 100 ms once files are indexed, with no files leaving the computer.

Visit PixaFind

Stack

  • Text embedding models
  • Image-text embeddings
  • Local vector index
  • NVIDIA GPU with CPU fallback