Every LocHub project setting lives in Project Settings > Plugins > LocHub, grouped into the categories below. These settings are saved to your project's Config/DefaultEditor.ini and apply to everyone who opens the project (unlike the per-user Node.js Executable setting at the end of this page).
Project Settings > Plugins > LocHub
| Setting | Type | Default | What it does |
|---|---|---|---|
| Service Port | Integer (1024–65535) | 47810 |
The local port the LocHub service listens on. |
| Auto Start Service | Bool | On | Starts the service (node <plugin>/Resources/LocHubService/lochub_service.mjs serve) automatically whenever Push, Pull or Open LocHub finds nothing answering on the port. |
| Release Policy | Validated / Approved Only | Validated | Sent to the service when it starts, and controls which strings a Pull may write into your archives. A running service keeps whatever policy it started with until you use Tools > LocHub > Restart Service. |
| Setting | Type | Default | What it does |
|---|---|---|---|
| AI Provider | Anthropic / OpenAI / xAI (Grok) / DeepSeek / Google Gemini / Custom (OpenAI-compatible) | Anthropic | Which provider translation and judging jobs use. Applied to the running service immediately, or — while a job is in progress — right after that job finishes. |
| Project Brief | Multi-line text | Empty | What the game is, its setting and tone, who the player is, and anything else a translator should know before touching a single string. Shared by every culture, sent with every translation and review request. Applied to the running service the same way as AI Provider: immediately, or right after a running job finishes. |
| Anthropic Auth | API Key / Claude Subscription | API Key | How LocHub authenticates with Anthropic when Anthropic is selected. API Key: enter it in the API Key field directly below. Claude Subscription: runs your own signed-in Claude Code CLI (claude) instead — hides the API Key field entirely, since no key is read or sent. Only shown when AI Provider is Anthropic. See AI Providers and API Keys for setup and limits. |
| API Key | Password-masked text | Empty | The selected provider's key — shown directly under AI Provider (and, for Anthropic, under Anthropic Auth, only when Anthropic Auth is API Key). One key per provider; switching AI Provider shows that provider's own key field and hides the others. Masked while typing; optional for Custom (OpenAI-compatible), where an empty key sends no key at all. Saved in Config/DefaultEditor.ini with the other LocHub settings, so it travels with the project like any other project setting. Applied to the running service the same way as AI Provider: immediately, or right after a running job finishes. |
| Anthropic Models | Translate Model / Judge Model | claude-opus-5-5 / claude-sonnet-5 |
The Claude model IDs used to translate and to judge. Only shown when AI Provider is Anthropic. |
| OpenAI Models | Translate Model / Judge Model | gpt-6-sol / gpt-6-luna |
The OpenAI model IDs used to translate and to judge. Only shown when AI Provider is OpenAI. |
| xAI Models | Translate Model / Judge Model | grok-4.7 / grok-4.7 |
The xAI model IDs used to translate and to judge. Only shown when AI Provider is xAI. |
| DeepSeek Models | Translate Model / Judge Model | deepseek-v4-pro / deepseek-flash |
The DeepSeek model IDs used to translate and to judge. Only shown when AI Provider is DeepSeek. |
| Gemini Models | Translate Model / Judge Model | gemini-3.8-flash / gemini-3.5-flash-lite |
The Google Gemini model IDs used to translate and to judge. Only shown when AI Provider is Gemini. |
| Base URL | Text | Empty | Custom only. The OpenAI-compatible API address, for example http://localhost:11434/v1; must start with http:// or https://. LocHub adds /chat/completions and /models. Logs show only its scheme, host and port. LocHub never sends an output-length cap to this endpoint — the server's own default applies. Sent to the service in an environment variable, never on its command line. A redirect from this address is refused, not followed. |
| Key Header | Authorization: Bearer / api-key | Authorization: Bearer | Custom only. How the API key is sent; api-key is for Azure OpenAI. |
| Custom Models | Translate Model / Judge Model | Empty / Empty | Custom only. Model IDs exactly as the endpoint lists them. Translate Model is required — without it LocHub does not start the service; an empty Judge Model uses the Translate Model. |
| Structured Output | JSON Schema / JSON Object / Prompt Only | JSON Schema | Custom only. How LocHub asks for JSON: try JSON Schema first, then JSON Object, then Prompt Only. |
| Input Price (USD per 1M tokens) | Number (0 or more) | 0 |
Custom only. Price of input tokens, for both models. |
| Output Price (USD per 1M tokens) | Number (0 or more) | 0 |
Custom only. Price of output tokens, for both models. With both prices at 0 the estimate shows no cost and Max USD cannot limit spending. |
| Max Parallel Requests | Integer (1–32) | 2 |
Custom only. The most requests LocHub sends to the endpoint at once, across every running job (not just one job's own requests). Every value in this range takes effect. |
| Request Timeout (seconds) | Integer (30–300) | 300 |
Custom only. How long one attempt may take before LocHub gives up on it; 300 is the most Node.js ever waits for an answer to start. A timed-out request is not retried as is — LocHub splits its group into smaller pieces instead, which is how a slow model still finishes. |
Each Models setting is a pair of free-text model ID fields (Translate Model, Judge Model) — you can point either one at any model ID your provider's API accepts, not just the defaults above. See AI Providers and API Keys for the full picture (cost estimates, key setup, Custom endpoint recipes). Every Custom setting is applied to the running service the same way as AI Provider: immediately, or right after a running job finishes.
| Setting | Type | Default | What it does |
|---|---|---|---|
| Ui Source Patterns | Array of wildcard paths | */UI/*, */Hud/*, */Widgets/*, */Menus/*, */WBP_* |
A string whose manifest source location matches any of these wildcards is sent to LocHub as widget text. |
| Coverage Exclude Patterns | Array of wildcard paths | Source/*Editor/*, */Tests/*, */ThirdParty/*, *.generated.h |
Paths, relative to the project folder, that the Coverage report skips. |
| Setting | Type | Default | What it does |
|---|---|---|---|
| Glyph Check Fonts | Array of Font asset references | Empty | Every listed font must have a glyph for every character of a translation, or Pull rejects it. An empty list turns the glyph check off. |
| Glyph Check Font Files | Array of font files (.ttf, .otf) |
Empty | Font files outside the asset system (for example RmlUi fonts) checked the same way as Glyph Check Fonts. |
| Write Dev Notes To Assets | Bool | On | Writes an answered translator question into the DevNotes of the source text asset. When off, every answer instead goes to Saved/LocHub/DevNotesProposals.md. Requires UE 5.8 or later — on 5.6 and 5.7 answers stay in LocHub and still reach the translator, they are just not written onto the asset. |
| Setting | Type | Default | What it does |
|---|---|---|---|
| Setup Foreign Cultures | Array of culture codes | de, fr, es, ja |
The foreign cultures Tools > LocHub > Set Up Localization Target adds to the Game target. Every run adds whichever listed cultures the target does not already have (a union; existing cultures are never removed) — so a culture you removed in the Localization Dashboard comes back on the next Set Up if it is still listed here. To drop a culture for good, remove it from this setting too. Unknown culture names (not one the engine recognizes) are skipped and named in the Set Up summary. |
Flags translations that are likely too long for your UI — German or Russian text often runs 30–40% longer than English — and tells the AI translator each string's limit up front, so it aims for a translation that fits.
Project Settings > Plugins > LocHub > Length Check
| Setting | Type | Default | What it does |
|---|---|---|---|
| Enable Length Check | Bool | On | Turns the check on. The other Length Check settings are greyed out while it is off. |
| Length Check Scope | UI strings / All strings | UI strings | UI strings checks only strings sent as widget text — the ones matching Ui Source Patterns (see Push above). All strings checks every string. |
| Max Length Ratio | Float (1.0–5.0) | 1.3 |
How many times as long as the English source a translation may be. Two decimal places are used. |
| Extra Characters | Integer (0–100) | 4 |
Characters allowed on top of the ratio, so a very short string ("OK", "Back") is not flagged for a few extra letters. |
| Culture Ratio Overrides | Map of culture → ratio | Empty | A different ratio for one culture (pt-BR) or a whole language (de, used for every German culture). An exact culture wins over its language; anything not listed uses Max Length Ratio. Ratios outside 1.0–5.0 are clamped; a key that is not a culture or language code is skipped and named in the Output Log. |
| Tell the Translator | Bool | On | Sends each string's limit with the translation request, so the AI aims for a translation that fits. |
| Length Severity | Warning / Must Confirm | Warning | Warning: a translation over its limit is flagged (band Y) and can be approved as usual. Must Confirm: approving or saving it needs Approve anyway / Save anyway, and an AI draft over the limit is written Needs fix — which the next job picks up and translates again. |
How the limit is computed. Limit = the English source's visible length × the ratio for the culture, rounded up, plus Extra Characters. Visible length counts what the player sees: format arguments such as {0} or {PlayerName} and rich-text tags count 0; a |plural(...), |ordinal(...), |gender(...) or |hpp(...) argument counts as its longest form; Chinese, Japanese and Korean characters, fullwidth forms, and emoji in the blocks U+1F300–U+1F64F and U+1F900–U+1F9FF (most faces, gestures, animals, food and objects — each skin-tone modifier counts 2 as well) count 2, while other symbols and emoji (for example the rocket, U+1F680, or the sun, U+2600) count 1; an accent or other mark that draws no glyph of its own, and a zero-width character, count 0 — a spacing mark (for example a Devanagari vowel sign) draws its own glyph and counts like ordinary text; the four rich-text entities & < > " count 1 each, the single character they stand for. A string whose source has nothing visible gets no limit. Example: "Save & Quit" has 11 visible characters; with the defaults its limit is 11 × 1.3 = 14.3 → 15, plus 4 = 19.