AI that runs on your computer, not ours
Answers are written by an open model running locally through llama.cpp. Your documents and questions never leave the device.
Two ways to use Pramvex
Search-only mode: no model needed
Install the app, add a folder, and you can search straight away: exact phrases, keyword passages, counts and file lists. Nothing else needs to be downloaded.
Local AI mode: optional
Open Settings › Models to add written answers. Pramvex looks at your computer’s memory and recommends a model. Nothing downloads until you click Download. Each file comes from a pinned release and is checked against a published SHA-256 before it is used. Pramvex then starts and stops the model server itself, listening only on your own computer (127.0.0.1).
Models offered
| Model | Licence | Download | Use |
|---|---|---|---|
| Phi-4-mini Instruct | MIT | about 2.5 GB | Recommended under 16 GB of memory |
| Qwen2.5 7B Instruct | Apache-2.0 | about 4.7 GB | Default for 16–32 GB |
| Qwen2.5 14B / 32B Instruct | Apache-2.0 | 9–20 GB | Advanced option on 48 GB or more |
| Nomic Embed Text v1.5 | Apache-2.0 | about 150 MB | Meaning-based search |
Sizes are approximate. The 14B and 32B models are offered as advanced options and have had less testing.
Requirements
Windows 11 on x64, macOS on Apple Silicon or Intel, or 64-bit Linux with WebKitGTK 4.1 (Ubuntu 22.04+, Debian 12+, Fedora 38+ or equivalent). Larger models give better answers but need more memory and run slower. Final minimum requirements will be published with the first public release.
Find the answer. Verify the evidence. Keep your files private.
Free to start. No account, no upload, no subscription.