tanakh/plamo-translate-desktop

Overview

PLaMo Translate Desktop is a Rust desktop application built with GPUI that runs the PLaMo 3 Translate model locally, so translation works without sending text to a server once a model has been downloaded. It translates between 53 languages, with Japanese ⇄ English as the default pair. Typical users are people who want private, offline-capable translation on Windows, Linux, or Apple Silicon macOS, including long documents and whole web pages.

The app ships no model weights. Model files are downloaded from Preferred Networks' Hugging Face repository, and each model shows the quality PFN publishes plus the gap from the unquantized version. Downloads resume after interruption and are verified by SHA-256. Long inputs are split into paragraphs, with earlier pairs passed as history for consistent wording; repetition loops are detected and retried in smaller chunks. It runs on CPU (AVX2/AVX-512) or GPU via Vulkan, and Linux source builds can also use ROCm.

A bundled browser extension for Chrome and Edge 137+ can translate selected text and entire pages using the app's model, or run standalone in the browser via wllama with the smaller 0.8B and 2B models.

Getting Started

1. Download a package for Windows (x86_64), Linux (x86_64), or macOS (Apple Silicon) from Releases and check it against SHA256SUMS.txt.
2. Windows: unzip and run plamo-translate-desktop.exe, keeping the DLLs beside it. Linux: extract and run plamo-translate-desktop (Ubuntu 24.04 build; needs glibc 2.39+, libgomp1, libxkbcommon-x11-0, libxcb1, and a CJK font). macOS: open the .dmg and drag the app to Applications.
3. Open the model button in the status bar, accept the PLaMo Community License, and download a model (0.8B, 2B, 8B, or 31B GGUF).
4. Paste or type text in the input pane. Translation streams in as it is generated; press Ctrl+Enter (Cmd+Enter on macOS) to translate immediately.

For the browser extension, open Settings in the app, choose Connect under Browser extension, then load the browser-extension folder via chrome://extensions in Developer mode. If you skip the app, the separate -browser-extension package translates inside the browser, but only with 0.8B or 2B models.

Constraints

The app is Apache-2.0 or MIT licensed, but the models are not in the repository and fall under the PLaMo Community License: personal and non-commercial use is free; commercial or organizational use requires registration when annual revenue is ¥1 billion or less, and a separate license from Preferred Networks above that. Published translations must indicate they are PLaMo output.

The macOS build is not notarized, so the first launch needs either the xattr -dr com.apple.quarantine command or Open Anyway in System Settings. The Intel NPU build is experimental and needs NPU selected manually as the device; it targets lower power use rather than speed. The app and the extension's native host each load the model, so both together keep it in memory twice. Browsers installed as Snap or Flatpak on Linux cannot start the host.

HotRepo's take

The project appeared on the Rust new-project list on 2026-11-09 with 103 stars, 7 forks, and a reported gain of 3 stars over a roughly 90-minute window starting 2026-11-09 11:15 UTC. There is no all-language daily rank, and the observation window is too short to support a trend claim. What stands out from the technical material is scope rather than raw popularity: a full local-model desktop client with quantization choices, resumable downloads, CPU/GPU/NPU device selection, and a browser extension that shares the app's model. Those are substantial implementation details for an early project, and they make it worth watching beyond the small star count.