Settings

Stored in this browser only. Keys are never displayed once saved: only the last four characters are shown.

Groq API access

Whisper transcription and LLM marking run on Groq.

In direct mode, any provider without a key in this browser is automatically routed through proxy.php, so keys can also live in config.php on the server.
console.groq.com, API keys.
Put keys in config.php on the server (see config.sample.php).

AI models

After a successful test the assessment list shows the models available to your account.

0 is most consistent.

Speech (radio messages and injects)

Groq's Orpheus voices are American only and capped at 200 characters. For a natural UK accent use ElevenLabs or OpenAI.

On Windows, Edge provides the natural "Microsoft Ryan / Sonia Online" en-GB voices. Chrome uses Google UK English.
Leave empty to use the key in config.php on the server.
Shared with OpenAI image generation.

Scene image generation

Nano Banana (Google Gemini image models) is recommended for photorealistic UK scenes. Groq and ElevenLabs have no image models.

Google AI Studio key (aistudio.google.com). Leave empty to use GEMINI_API_KEY in config.php.
Same key as the speech section.
Must accept POST /images/generations with an OpenAI-style body and return b64_json or url.

Assessment and player

Appearance

The dark workbench suits the builder; the light theme is easier on trainees during an exam.

Marking guidance

Injected into every assessment prompt: service SOPs, terminology, expectations.

Access password

Every page shows a login screen until the access password is entered. Changing it here applies to this browser; the default applies elsewhere until changed there too.

At least 8 characters.

The built-in login works on any host, including exported packages, but it is a client-side gate. For server-enforced protection on Apache or XAMPP, enable the Basic Auth block in .htaccess (a ready .htpasswd is included).

Data

Scenarios, images and attempts live in this browser's IndexedDB.

Calculating storage

Backups include settings but never include API keys.

Security and governance notes

Data flow. Audio goes to Groq for transcription; transcripts and scenario text go to Groq for marking; radio text goes to your chosen speech provider; image prompts go to your image provider. Nothing is sent anywhere else. Check each provider's data-processing terms against your DPIA before live use.

Human in the loop. AI marking is a recommendation. An authorised examiner must review and confirm every grade, consistent with the service's AI-Assisted Marking Tool disclaimer.

Keys. Prefer config.php on the server (proxy mode or automatic fallback). Keys entered here sit in this browser's local storage and are only ever displayed masked.