Goblin Forge — Game Localization Workbench

Forge your game's text into any language.

Goblin Forge is a localization workbench built specifically for game text — for you, your team, and whatever AI goblin you choose to put to work — giving you a complete forge for language work. (And if you'd like to use it on other kinds of text, the forge isn't picky — go right ahead.)

Goblin Forge runs fully offline and never forces AI on you. If you do want to use AI, it also gives you the tools to rein in AI translation and make it more consistent.

Every feature comes straight out of the maker's own painful experience doing game translation, paired with hands-on experience using AI.

You can polish translations entirely by hand, or let AI assist the whole way — the full core toolset works out of the box, with no reliance on any online service.

That said, cloud models do need a connection — remember to bring your own API key. Your API key is stored locally and is never uploaded to Goblin Forge servers. Keep your own machine secure, and your key stays under your control.

Pure offline, by-hand mode — the forge gives you:

• Project Library Manager: see real-time project progress, manage your asset libraries (project + global), and run batch operations.
  
• Translation Editor: source and translation side by side, context always in view, key terms and code auto-highlighted and protected so they don't get broken by accident. Live glossary and translation-memory match hints. Switch anytime between raw view (shows the code) and card view (renders the line breaks). Built-in find & replace, plus auto-wrap (basic, based on source length).
  
• Translation Memory: translated lines are saved automatically, scored by source, with usage counts — click to jump back to where it came from in the editor. Add entries by hand, or import external memory.
  
• Term Workbench: global and project libraries side by side, push or pull with one click. Supports multiple translations per term, auto-detects conflicts, and digs term candidates out of the source by word frequency.
  
• Character System: tag each line with its speaker, with tone and pronoun references.
  
• Placeholder protection / regex code protection: keeps the variables and code in your game text from getting broken by accident.
  
• Scan Engine: automatically checks placeholder and term integrity, keeping game-breaking errors out before you export.

Connect AI, and on top of everything above, your translating gets a boost.

You'll need a local or cloud LLM set up (Ollama free & local / Gemini / OpenAI / Anthropic / DeepSeek / Kimi, or any OpenAI-compatible endpoint — remember to bring your own API key; the forge doesn't ship one):

• AI translation & AI review (early): glossary, characters, and protection rules are injected automatically — no mindless literal translation.
  
• Two model tiers: economy and premium — use different model versions depending on content difficulty (you decide which; automatic switching isn't supported yet) to cut costs.
  
• Batch translation: automatically skips already-translated / locked lines, and on rate limits it backs off and retries.
  
• Goblin AI Assistant: always aware of your project progress; ask it all kinds of translation questions — including how to use the software and what a given feature does. You can even just chat with it (not recommended — it costs money), and it can summarize and jot down notes, helping users with a short working memory keep track of what they've done. It can be given its own model, and can run on a local model to save costs.
  
• Cost monitoring (estimated): it estimates before a batch, tallies spend in real time (you enter the AI's pricing yourself), and stops automatically at the limit you set. (Note: this is an estimate, so it may differ from the real price — check your model's own dashboard now and then.)

Import & export: same structure in, same structure out

The forge imports CSV and exports CSV / JSON, currently in UTF-8, with more formats and encodings coming. Everything you import keeps its structure on export:

• All protected code and placeholders are output unchanged.
  
• However many columns you import, that's how many you export.
  
• In short: the structure you put in is the structure you get back — only the text has been translated.

The forge supports atomic saving and automatic backups to reduce the risk of file corruption from software crashes, system crashes, or sudden power loss — though I'd still personally suggest saving by hand now and then.

In Early Access, beyond the above, coming soon:

• Team merge (splitting already available, with SHA256): for smoother multi-person collaboration, especially spotting and unifying translation conflicts. ~50% done.
  
• Polishing, review (advanced), and fixing: polish by hand, or use AI for unified passes. ~80% done.
  
• Furigana (Ruby) handling for Japanese (experimental): mainly to handle Japanese phonetic annotation; for other languages it can be used for puns — "written A, read B." ~90% done, still needs testing.

On the roadmap, still to come:

• Visual QA (preview layout/overflow in a mock dialogue box): tune text length and line breaks before importing into the game, and preview whether your chosen font can actually display the language's characters — instead of a row of □□ (a.k.a. missing glyphs / tofu).
  
• Smart auto-wrap (advanced): break lines based on the source's phrasing and dialogue or the text window, with AI assistance, so the break points match each language's conventions.
  
• Goblin evolution (memory & automation): let the goblin assistant remember your habits and preferences — and if you'd rather it didn't, you can switch it off — plus more automation driven by user feedback.

The forge is a complete one-time purchase: fully offline-capable, fully usable by hand, AI-assisted all the way, with your own API key. No subscription, no servers — your text always stays on your own machine.

This forge is largely designed from one maker's real workflow and hard-earned lessons, so some edges may still need refining. Please send feedback as you go — I'll keep improving and adjusting it.