🇫🇷 Lire en français

Hoshi2Star ★

星 → ★ — CAT editor + LLM orchestrator for fan translation of Japanese RPG games

Screenshots

Before & After

Original (日本語)	Translated (English)

Hoshi2Star in action

1. Open the app

2. Load a game — segments extracted

3. Configure Ollama and translate

4. Segments translated in 27s — QA score 100

---

Features

• Open and extract RPG Maker MV/MZ games
• LLM-assisted translation (local Ollama, no API key needed)
• Cross-project Translation Memory (TM) with exact lookup
• Automatic QA: placeholders, line length, UTF-8 BOM
• Export translations back into the game files
• 3-panel CAT interface: Files | Grid | TM + QA
• Project list — recent projects with progress cards, continue or delete with one click
• Per-segment translate button — retranslate a single segment without batch
• Glossary auto-extraction — LLM detects key terms automatically on project open
• TM fuzzy matching — 80% similarity threshold with Levenshtein distance
• Export QA report as standalone HTML

---

Supported Engines

Engine	Status	Formats
RPG Maker MV	✅ Supported	.json, .rpgmvp
RPG Maker MZ	✅ Supported	.json, .rpgmvp
RPG Maker VX Ace	🔜 F3	.rvdata2
Wolf RPG	⚠️ Partial	.dat
RPG Developer Bakin	🔜 F5	.rbpack

---

Prerequisites

• Ollama installed: https://ollama.ai
• Recommended model: ollama pull qwen3:4b-instruct-2507-q8_0

The -instruct variant responds directly without a thinking phase, which produces faster and more reliable translations.

• Linux: webkit2gtk-4.1 (usually already installed)
• Windows: no additional prerequisites

---

Cloud LLM with RunPod (optional)

Instead of running Ollama locally, you can use a cloud GPU on RunPod for faster translation with larger models.

Setup

1. Create a pod on RunPod (recommended: RTX 4090 or A40)
2. Set these Environment Variables:

   OLLAMA_HOST   = 0.0.0.0
   OLLAMA_MODELS = /workspace/ollama-models
   

3. Expose HTTP port 11434
4. Set this Start Command (replace the model name if needed):

   bash -c "mkdir -p /workspace/ollama-models && apt-get update && apt-get install -y zstd && curl -fsSL https://ollama.com/install.sh | sh && ollama serve & until ollama list > /dev/null 2>&1; do sleep 1; done && ollama pull qwen3:4b-instruct-2507-q8_0 && wait"

   until ollama list polls every second until the server is ready before pulling — more reliable than a fixed sleep 5.

5. Once the pod is running, copy the proxy URL from RunPod → Connect:

   https://[POD_ID]-11434.proxy.runpod.net
   

6. Paste this URL in Hoshi2Star Settings → Ollama URL and click Test

Recommended batch sizes

GPU	VRAM	Batch size
RTX 4090	24 GB	30–40
A40	48 GB	40–50
A100	80 GB	50–60

Remember to stop your pod after translation — RunPod charges per minute. Models stored in /workspace persist across restarts (Volume disk), so you only download them once.

---

Installation

Linux:

chmod +x hoshi2star_*.AppImage
./hoshi2star_*.AppImage

Windows: download and run the .msi from GitHub Releases.

---

Quick Start

1. Start Ollama: ollama serve
2. Open Hoshi2Star
3. Click "Open Game" → select the game folder
4. Select a file in the left panel
5. Click "Translate" → configure Ollama (URL + model)
6. Start translation
7. Review and edit segments in the grid
8. Click "Export" to apply translations to the game

---

Development

Prerequisites: Rust stable (rustup), Node.js LTS + pnpm

Linux extra: webkit2gtk-4.1, base-devel

git clone https://github.com/KATBlackCoder/Hoshi2Star
cd Hoshi2Star
pnpm install
pnpm tauri dev

Tests:

cargo test --manifest-path src-tauri/Cargo.toml
pnpm typecheck

---

Tech Stack

Layer	Technology
Desktop runtime	Tauri v2
Backend	Rust, sqlx, tokio
Frontend	React 19, TypeScript
UI	shadcn/ui, TanStack Table v8
State	Zustand
Database	SQLite (embedded)
LLM	Ollama (local)

---

Roadmap

See ROADMAP.md for the full development plan.

---

License

MIT — see LICENSE