Diplo Translator Omni uses Umbraco.AI for all model requests. Configure Umbraco.AI in the host website first; Omni does not store, create or manage provider credentials.
1. Install Umbraco.AI and a chat provider
Follow Umbraco’s Getting Started guide to install Umbraco.AI, then install a provider integration that supports Chat for your Umbraco version. The available providers and their setup steps evolve independently, so use the provider-specific instructions linked from the Umbraco.AI documentation rather than copying an old package or configuration example.
Restart the website after installation and confirm that the provider is visible and healthy in the Umbraco backoffice before configuring Omni.
Marketplace quick links
- Umbraco.AI — the core package.
- All official Umbraco 17 AI packages.
- Providers: OpenAI, Anthropic, Google, Microsoft Foundry, Mistral, DeepSeek, Hugging Face, Amazon, Together AI and Fireworks AI.
2. Create a chat profile for translation
In Umbraco.AI, create a Chat profile that selects the provider, model and behaviour to use for translation. Profiles are reusable, so a profile is the right place to centralise a model choice, system instructions and guardrails.
- Use a clear, stable alias, such as
translation-qualityortranslation-fast. - Keep temperature low for consistent, literal output.
- Allow enough output tokens for the longest field you expect to translate.
- Test with representative rich text, links, placeholders, product names and every target language.
- Use provider or Umbraco.AI guardrails where your organisation requires them, but make sure they do not reject normal editorial language.
Omni supplies a translation instruction with each request. A profile-level instruction is optional; if you add one, keep it short and avoid competing instructions such as “write marketing copy” or “summarise”.
Model recommendations
There is no universally best translation model. Start with a current general-purpose chat model from an approved provider, then compare profiles using your real content before enabling one for editors.
- Everyday website translation: choose a capable, current general-purpose model with strong multilingual support and a context window large enough for your longest rich-text values.
- High-value or regulated content: favour the model that performs best in a blind review by native-language editors; use a separate quality profile and require review before publishing.
- Large-volume, low-risk content: use a lower-cost model only after testing terminology, markup preservation and target-language quality. Keep a higher-quality profile available for corrections.
Do not choose solely by benchmark, price or model name. Translation quality varies by language pair, subject matter and the provider’s current model version. Pin a tested model version where the provider supports it, and repeat a small acceptance test when changing models.
API keys and provider accounts
Create a dedicated project or service credential with your selected provider, enable billing or usage limits where required, and then enter the key using the provider integration’s documented configuration method. Provider API costs are separate from the free Umbraco packages and from the Omni licence. Common starting points are OpenAI API keys, Anthropic API keys, Google AI Studio API keys and Mistral API keys.
Keep secrets out of source control and browser-delivered configuration. Use environment variables, the hosting platform’s secret store or another protected configuration mechanism, with separate credentials and spend limits for development, staging and production. Rotate and revoke a key immediately if it is exposed.
3. Make the profile available to Omni
- Sign in as an administrator and open Settings → Translator settings.
- Enable the live Umbraco.AI chat profile or profiles that translators may use.
- Select a default profile for normal jobs.
- Run a small translation job and review the saved target-language draft before giving editors wider access.
Alternatively, set DiploTranslatorOmni:ChatProfileAlias as a fallback. The administrator-managed allowed profiles and default take precedence. See AI profiles & prompts for Omni’s profile-selection rules.
Operational checklist
- Set a provider budget, rate limit or alert before opening translation to a wider editorial team.
- Review what content may be sent to the provider, especially personal, confidential or regulated information.
- Keep human review and normal publishing permissions in place; an AI translation job saves a draft, it does not replace editorial approval.
- Record the provider, model and profile alias used in your release or support notes so changes are traceable.