LinguaForge - Local AI Translation

# LinguaForge - Local AI Translation for Unity

**Translate your game locally. No cloud. No API costs. No data leaving your machine.**

LinguaForge is the first Unity localization tool powered entirely by local AI inference. Translate your String Table Collections using AI models running on your own hardware—no subscriptions, no per-character fees, and complete privacy for your unreleased content.

---

## ⚠️ REQUIREMENTS

- **Unity Localization** package (com.unity.localization) — sold separately

- **Ollama** installed locally (free from ollama.ai)

- **AI Model**: qwen2.5:7b-instruct recommended (~4.5GB)

---

## 🔑 KEY FEATURES

### 100% Local Processing

Your game text never leaves your machine. Perfect for unreleased titles, confidential projects, or studios with strict data policies. No cloud APIs, no data transmission, complete privacy.

### No Per-Word Costs

Unlike cloud translation services that charge per character, LinguaForge uses your local hardware. Translate unlimited text after your one-time purchase—no surprise bills, no usage caps.

### Works Completely Offline

Once Ollama and your model are set up, no internet connection is required. Translate on planes, in cafes, or anywhere you develop.

### Smart String Preservation

Automatically preserves Unity Smart String syntax including:

- Named variables: `{playerName}`

- Indexed placeholders: `{0}`, `{1}`

- Plural formatting: `{count:plural:# item|# items}`

- Choice formatting: `{gender:choose(m|f):He|She}`

- Nested placeholders

### Glossary System

Define terms that should never be translated:

- Brand names and trademarks

- Character names

- Game-specific terminology

- Technical terms

Categories help organize large glossaries.

### Translation Memory

Stores previous translations for instant reuse:

- Exact match lookup for repeated text

- Ensures consistency across your project

- LRU eviction keeps memory manageable

- Reduces AI calls and saves time

### Batch Translation

Translate entire collections with:

- Multi-language support in one operation

- Real-time progress tracking

- Estimated time remaining

- Cancel anytime with partial results saved

- "Missing only" mode skips existing translations

### Quality Validation

Automatic checks ensure translation quality:

- Placeholder preservation verification

- Character limit warnings

- Length ratio anomaly detection

- HTML/Rich text tag preservation

---

## 🖥️ INTERFACE

**Five dedicated panels:**

1. **Translation** — Single entry translation with collection browser

2. **Batch** — Bulk translate multiple collections and languages

3. **Glossary** — Manage Do-Not-Translate terms and translations

4. **Memory** — View, search, and prune translation memory

5. **Settings** — Configure Ollama connection, model, and quality options

Modern UI Toolkit interface with OTG Studio's signature dark theme and green accents.

---

## ⚙️ REQUIREMENTS

**Software:**

- Unity 2021.3 LTS or newer (Unity 6 supported)

- Unity Localization package 1.4+

- Ollama (free from ollama.ai)

**Hardware:**

- 8GB+ RAM recommended

- 4GB+ GPU VRAM for best performance

- 5GB storage for AI models

**Recommended Model:**

- qwen2.5:7b-instruct (~4.5GB, best quality/speed balance)

- qwen2.5:3b-instruct (~2GB, faster, good quality)

---

## 📊 PERFORMANCE

- Single entry: 2-5 seconds typical

- Batch of 100 entries: ~5 minutes

- Translation Memory hits: Instant

- 214 unit tests passing

- Automatic placeholder repair

---

## 📚 DOCUMENTATION

Includes comprehensive documentation:

- Quick Start Guide (5 minutes to first translation)

- User Guide (complete feature documentation)

- Developer Guide (API reference, extension points)

- Troubleshooting Guide

---

## 🛠️ SUPPORT

- **Documentation:** Included in package

- **Discord:** discord.gg/nJqDGWmw

- **Email:** contact@otgstudio.com

---

## 🏢 FROM OTG STUDIO

We build AI-powered game dev tools that run locally—no cloud, no API fees, no data leaving your machine.

**GAME DEV | AI | TOOLING**

Also check out **BehaviorForge** — AI-powered behavior tree generation for NodeCanvas.