Dhito is coming to Windows.
We are actively adapting our local Whisper, Florence-2, and BGE-small-en-v1.5 embedding pipelines to Windows 11. Get private, on-device semantic search for your PC.
Privacy-first product. Your email is only used for beta updates and is never shared.
Why Dhito for Windows?
100% On-Device Privacy
No document uploads or cloud vectors. Index files, transcribe audio, and process queries fully locally.
Direct PC Hardware Acceleration
We optimize our models using local GPU/NPU engines to run with maximum speed and zero performance drag on your PC.
Looking for the Mac version?
Dhito for macOS is available to download and try immediately.
What Dhito for Windows will do
The same thing the Mac version does: index your files locally and let you find them by describing what they are about instead of recalling what they were named. A local embedding model matches your query on meaning, so searching for a policy on remote work surfaces the handbook even though the two share no words.
That extends past documents. Images are indexed by their visual contents through a local vision model, so an untitled screenshot is findable by what is in it. Audio and video are transcribed on-device with Whisper and indexed by meaning, with results pointing at the timestamp where something was said rather than at the file as a whole. And once a document is found, a local language model can answer questions about it and cite the page the answer came from.
Why macOS came first
Apple Silicon ships one memory pool shared by the CPU, GPU, and Neural Engine. A model can read a large index without copying it across a bus first, which is what makes it practical to run Whisper, Florence-2, and an embedding model in the background on a laptop without the machine becoming unusable. Building against a single, predictable architecture let us find out what on-device search actually needs before committing to a second platform.
Windows hardware is more varied — discrete GPUs, integrated graphics, and NPUs that differ by vendor — so the work in progress is largely about making the same pipelines perform well across that range rather than only on the fastest configurations. The waitlist is where beta invites go out as those builds become usable.
How this differs from Windows Search
Windows Search indexes filenames and file contents as literal text, which is the same lexical approach macOS Spotlight takes and it fails the same way: if your words are not the file's words, nothing comes back. It also has no way to look inside an image or listen to a recording, so the media on your drive is matched on its filename alone.
Semantic search removes the guessing. You describe the thing you are looking for, and matching happens on meaning — across documents, images, and recordings at once, with everything staying on the PC.