Bring your own AI provider

Automatic Translation and AI Models

Connect a supported provider from your own WordPress site, choose Standard or Enhanced translation, and run protected multilingual jobs without routing your API key or model requests through Stag at Last.

Before the first request

Prepare access, scope, and one safe test.

AI model settings and scoring are available in StagLingo Free. Running the built-in Automatic Translation workflow requires StagLingo Premium and uses the account, model access, and paid usage supplied by your chosen provider.

Provider account

Create an API key

Use a provider account you control. A ChatGPT, Gemini, or Claude consumer subscription does not necessarily include API credits.

Model access

Confirm the exact model ID

Copy an exact model identifier enabled for your API account. StagLingo does not silently replace that choice during an update.

WordPress environment

Check Diagnostics

Verify TranslatePress languages, storage access, PHP requirements, and the other checks reported by StagLingo before a live run.

Publishing control

Protect unfinished URLs

Use Index Control to keep incomplete language versions noindex, follow until translation and review are complete.

Initial scope

Start with one real page

Choose a representative published page and a small language set before scheduling a broader production batch.

Cost control

Check provider limits

Review token limits, rate limits, account balance, and provider data controls. Model calls are billed by the provider, not StagLingo.

If the plugin or license is not ready yet, complete Getting Started with StagLingo and review Plans, Licenses, and My Account.
AI Models and API

Connect and test one provider.

Open Tools → StagLingo → AI Models. Save one provider configuration, then run the minimal connection test before attempting a translation or scoring job.

  1. 01

    Select OpenAI, Google Gemini, Anthropic Claude, or an OpenAI-compatible endpoint.

  2. 02

    Enter the exact Model ID available to your account.

  3. 03

    Paste the API key. For a compatible provider, also enter its required HTTPS endpoint.

  4. 04

    Keep the default 90-second timeout and 12,000 output-token limit unless your provider or workload requires a deliberate change.

  5. 05

    Select Save AI model settings, then Test connection.

  6. 06

    Continue only after StagLingo reports AI connection verified.

StagLingo AI Models and API settings for selecting a provider, model, and API key
Custom endpoints must use HTTPS. The timeout accepts 20–180 seconds and the maximum output setting accepts 1,000–32,000 tokens. Higher limits do not override restrictions imposed by the selected model or provider account.
Credentials and data

Requests travel from your site to your provider.

Automatic translation and scoring send selected page text, target-locale context, and required structure directly from the WordPress installation to the provider you configured. Stag at Last is not a proxy for these requests.

01

Stored in WordPress

Your saved API key is kept in a non-autoloaded WordPress option and encrypted with a key derived from the installation's WordPress authentication salts.

02

Sent to your provider

The site uses the credential only to authenticate direct model requests. The selected provider receives the submitted translation or scoring content.

03

Not sent to Stag at Last

The plugin vendor does not receive the API key or route the model request. Keys are excluded from StagLingo exports, logs, reports, and learned rules.

StagLingo prefers AES-256-GCM through OpenSSL and can use Sodium secret-box as a fallback. If neither encryption method is available, it refuses to save a new key. Your provider's terms, retention policy, regional processing, and account controls still apply. See Data Handling and Security for the complete boundary.
Choose the review depth

Standard creates the first result. Enhanced reviews the rendered result.

Both modes use StagLingo's protected export and import foundation. Enhanced is not just a longer first prompt: it writes round one, fetches the actual translated front end, and then creates targeted second-round corrections.

Standard

Automated first-round workflow

Use Standard when you need a protected initial translation that you will review separately or when provider usage needs to stay lower.

  • Creates an immutable source snapshot.
  • Runs one locale-specific translation and review pass.
  • Validates and writes through the protected Import core.
  • Loads the latest applicable abstract quality rules.
  • Preserves existing non-empty translations by default.
For more visible package-by-package control, use the Manual Two-Round Translation Workflow. For later maintenance of pages that already contain translations, use the Incremental Translation Workflow.
StagLingo Automatic Translation screen for selecting a page, target languages, and quality mode
Automatic Translation

Schedule one published page and the locales it needs.

Open Tools → StagLingo → Auto Translation. StagLingo reads the available target languages from TranslatePress and processes each locale independently, even when several languages are scheduled together.

  1. 01

    Enter one published default-language URL or content ID.

  2. 02

    Select the TranslatePress target languages for this job.

  3. 03

    Choose Standard or Enhanced.

  4. 04

    Leave replacement disabled to write only missing translations.

  5. 05

    Enable Also review and replace existing non-empty translations only when deliberate revision is required.

  6. 06

    Select Translate, validate and write, then keep the job page available while it advances.

Languages may be scheduled in groups of up to three, but each locale receives its own translation and review work. Enabling replacement broadens what the job may write; use it only after checking backups and the intended page scope.
Read the result correctly

Separate language quality from application evidence.

StagLingo's independent 10-point audit evaluates the translated language. Suspected front-end application or visibility problems are reported separately so a storage or cache issue does not artificially change the language score.

Language score

What the wording earns

Accuracy, completeness, native fluency, terminology, tone, and visible composition contribute to the language-quality result.

Application status

What the front end shows

Missing rendered text, stale output, or other suspected application failures are identified separately and should be verified on the live page.

Protected write

What StagLingo records

The import transaction includes a private backup, validation, storage read-back, History entry, and the cache-clearing or verification steps available to the site.

StagLingo Scoring screen showing an independent translation-quality audit
A score is evidence for review, not a ranking or indexing guarantee. Before making a language URL indexable, inspect the rendered page and follow Quality and Publishing Controls.
Safe stops and recovery

Resume the same snapshot, or stop when the source changed.

StagLingo retains a small set of recent private source snapshots so an interrupted locale or Enhanced second round can resume without rebuilding row identity. That safeguard deliberately refuses an unsafe resume after the source changes.

Connection test fails

Check the provider choice, exact model ID, API key, account balance, model access, and provider rate limits. A compatible custom endpoint must use https://.

Provider responds without valid structure

A network response alone is not enough. The connection test and translation workflow require the structured output expected by StagLingo.

Job stops partway through

Use the available resume action for the same saved job and immutable source snapshot. Completed locales should not need to restart.

Resume is blocked

If the current English source fingerprint differs from the saved snapshot, create a new job so translations are not applied against stale source identity.

Completed with warnings

Read the warning and inspect the translated front end. A warning can require verification without automatically meaning that every written translation failed.

Changes are not visible

Check the intended locale URL and clear relevant page, plugin, server, and CDN caches before concluding that the import did not apply.