Who this is for: administrators. If you author courses but don't administer the platform, you want translating in authoring instead—nothing on this screen is needed to translate a course.
Sidebar → ADMIN → Translation Settings
This screen controls how translation works across your whole platform: whether AI translation is available at all, which provider runs it, what it costs, and who gets told when a job finishes. It has three tabs.
AI Configuration (Not available yet)
Enable AI Translation—the master switch.
"Master switch for the AI translation method. When off, the AI option is hidden behind a 'Coming soon' tooltip on every course, and all AI translation API calls are rejected. Affects all clients."
Read that last sentence carefully. This is platform-wide, not per-client. Turning it off removes AI translation from every course belonging to every client on the platform, and rejects any API call already heading for a provider.
Choosing a provider
Default Provider—one of OpenAI, Anthropic, Gemini, DeepL, or Google Translate. This is the provider every translation uses unless a language-specific override says otherwise.
Default Model Override—pins a specific model. Not every provider supports this; when you select one that doesn't, the field explains itself ("Not used — this provider's API has no model selection").
Batch Size—how many segments are sent to the provider per request.
API Keys
One field per provider. "Keys are masked after saving. Enter a new value to replace."
You cannot read a key back once saved—you can only overwrite it. Keep your own record somewhere appropriate; this screen is not a password store.
AI translation will not work without a key for whichever provider you have selected, even with the master switch on.
Model Overrides
Sets a non-default model per provider for OpenAI, Anthropic, and Gemini. Leave blank unless you have a specific reason.
Pricing Overrides
"Override the built-in price list used for cost estimates (USD). Leave blank to use the shipped catalog. LLM providers are priced per 1M tokens; DeepL and Google Translate per 1M characters."
Poet ships with a price list and uses it to work out the cost shown against a completed translation. If your organization has negotiated different rates with a provider, enter them here so the figures match your actual bill.
Leave blank, and the shipped catalogue is used.
Language-Specific Providers
"Override which provider to use for specific languages."
Give a language code and a provider, and select Add. Useful where one provider is noticeably better for a particular language—you might run most languages through your default and route Japanese somewhere else.
Saving
Changes take effect when you select Save Configuration.
Translation Memory
A fuller view of the memory than the sidebar page offers for administrators who need to audit or bulk-edit it.
Per-language counts across the top—select a language tile to filter the entries below.
"Import—"Upload XLSX or JSON. XLSX columns: sourceText, targetLanguage, translatedText. JSON: array of objects with same fields."
The entries table, with filters, the sidebar page does not have:
- Search—across source text and translations
- Scope—All entries, this organization, or Global
- Target language and Source language
- Last updated—by date
- Sort
Each row shows the source, language, translation, and scope, with a delete control.
Scope is the column that matters. Global entries are shared across the entire platform; these organization entries belong to one client. Deleting or editing a global entry affects every client, so treat that as a platform change rather than a local fix.
For what Translation Memory is and how authors use it, see Translation Memory.
Notifications
Enable email notifications—"Send email notifications when translation workflow events occur."
Off by default.
This is the only setting in Poet that sends email about translations. If your team expects to be told when a translation finishes, this is what they are waiting on.
Related
- Translation Memory
- Translating in Authoring