WideBot Assistant

Settings

Speech recognition — what the citizen speaks Listed in the recognition engine's own spelling, because that is what gets sent to it.

Reply language — what the assistant speaks back Deliberately separate from the one above: an authority can take Egyptian and answer in formal MSA.

Voice Voices for the reply language are listed first; the rest stay selectable.

TTS engine

STT engine — Arabic The Arabic-specialised model is materially more accurate on Arabic than the multilingual one.

STT engine — everything else and auto-detect

Provider

Base URL (OpenAI-compatible only)

Model name

Credentials are configured on the server and are not editable here.

Repair transcripts When a correction changes the text, the original is shown beneath it as “heard:”, so a repair can never silently rewrite the citizen.

Test case Scopes the built-in script to one transaction. Ignored when a custom prompt is set.

System prompt A custom prompt replaces the built-in one rather than being appended to it — merging would leave you debugging text you cannot see. Only the reply-language and numbers-as-words lines are prepended, because the synthesiser needs them.

Welcome message Each language has its own. Edit one, pick another, edit that — all of them save together.

Show the built-in prompt

TTS…

STT…

LLM…

Mic…

Demonstration system — the transaction is simulated and no record is created or changed.

Admin

show empty visits

One block per turn: what was heard, what was repaired, what the model answered, and how long each step took. Newest first.

New account

A User holds conversations and can change language, voice, engines and prompt for their own account only. An Admin also sets the defaults everyone starts from, the LLM, and sees sessions, logs and accounts.