Gen3 Accessibility
- Open the account menu in the sidebar footer (click your name or profile icon).
- Choose Light mode or Dark mode to override the organization default for your signed-in account.
- In GT Chat, open Chat Window Preferences when you need a wider reading column, quieter tool activity, or icon-only message actions—see Chat Window Preferences.
- For speech in chat, confirm operators enabled Platform Multimodal Settings (STT/TTS) and your agent has Speech Settings configured—then use the microphone control or Read aloud on assistant messages.
- If voice features are missing or fail, work through the troubleshooting section below before opening a support ticket.
GT AI OS targets WCAG 2.1 AA–aligned semantics, keyboard reachability, and readable contrast across tenant workflows. Light and dark appearance, chat layout tuning, and optional speech input/output let each user adapt the product to their environment without changing deployment security or data isolation. Operators set organization defaults; users keep personal overrides.
- Click your name or profile icon in the sidebar footer.
- Select Light mode or Dark mode.
The choice applies immediately to the tenant app chrome, chat surfaces, and instructions UI. Your preference is saved to your user profile and persists across sign-in sessions on that account.
Tenant owners and operators set the default appearance on Control Panel → Settings → Tenant appearance:
- Light mode — Light neutral surfaces with the deployment accent color.
- Dark mode — Dark neutral surfaces; accent tokens stay the same.
New users and sessions without a personal choice start in this default. Per-user light/dark overrides take precedence over the organization default. Tenant appearance does not change Control Panel operator chrome or compliance frames.
Accent color (buttons, sidebar, chips) is configured in the same Tenant appearance section. Light and dark mode can use separate accent palettes—operators may customize dark accents independently or use Match dark accent to light.
Chat Window Preferences in GT Chat helps users who need:
- A Focused or Wide reading column instead of the default comfortable width
- Hide activity details for a calmer transcript during long agent runs
- Icons only message actions to reduce visual noise
- Disabled streaming when incremental token rendering is distracting
Open Chat Window Preferences from the GT Chat header toolbar. Platform defaults come from Control Panel Settings → Default chat window preferences; users reset to those defaults with Restore all chat defaults.
Speech features require deployment, platform, and agent configuration. All three layers must be enabled before microphone and read-aloud controls appear in chat.
| Layer | Where | What to verify |
|---|---|---|
| Platform multimodal | Control Panel → Models → Platform Multimodal Settings | Speech-to-text (STT) and Text-to-speech (TTS) enabled for the deployment |
| Default models | Control Panel → Models → Default Models | Default STT and Default TTS point to enabled catalog models |
| Agent speech settings | Building Agents → agent editor → Speech Settings tab | Agent STT/TTS model selections (visible when platform gates allow speech) |
When platform STT or TTS is disabled, related UX is hidden and matching API routes return a clear deployment message.
In GT Chat, open the microphone control near the composer to access voice options.
| Mode | Behavior |
|---|---|
| Normal (Speech to Text) | Records your speech, transcribes into the composer, and stops. You review and send manually. Always available when platform STT is enabled. |
| Full (Hands-Free Mode) | Listens, transcribes, sends your turn automatically after a pause, waits for the agent reply, and can read the reply aloud—continuing turn by turn until you turn it off. Available only when platform TTS is enabled and the agent has a resolvable TTS model. |
When TTS is disabled at the platform level, the microphone popup shows Speech to Text only—no hands-free entry path.
The popup shows current status (listening, transcribing, sending, speaking). Turn off Hands-Free Mode from the same control when you want to return to typed messages.
On completed assistant messages, Read aloud uses the agent's configured TTS model (or deployment default) to synthesize the visible answer. One-click read aloud and hands-free reply audio both respect the agent Speech Settings defaults saved for that agent.
Read aloud is separate from Hands-Free Mode—you can listen to a single message without enabling continuous voice conversation.
| Symptom | Likely cause | What to try |
|---|---|---|
| Microphone or Read aloud missing | Platform STT/TTS disabled | Ask an operator to enable capabilities on Platform Multimodal Settings and confirm default STT/TTS models |
| Control visible but action fails | Agent speech models unset or invalid | Edit the agent → Speech Settings; pick enabled STT/TTS models |
| Browser blocks recording | Microphone permission denied | Allow microphone access for the tenant app origin in browser settings |
| Transcription errors | STT provider unreachable or misconfigured | Operator checks model provider health on Models; retry after provider recovery |
| No audio on read aloud | TTS disabled or silent output device | Confirm TTS enabled at platform and agent level; check OS audio output |
| Hands-Free sends unexpectedly | Full voice mode active | Turn off Hands-Free Mode from the microphone control |
| No Hands-Free option in popup | Platform TTS disabled or agent TTS unset | Enable TTS on Platform Multimodal Settings, set default TTS, and configure agent Speech Settings |
Dataset audio ingestion and vision flows have separate platform gates—speech troubleshooting above applies to GT Chat voice and read-aloud only.
Tenant app surfaces use labeled controls, dialog roles where appropriate, and aria-live regions on chat transcripts during streaming. Prefer visible focus rings and browser defaults when navigating by keyboard. Report focus traps or unreachable controls through Getting Support with the route and browser version.
When your deployment uses Groq Orpheus v1 English for TTS, bracketed directions such as [cheerful] or [whisper] in the assistant message text control how Read aloud sounds. The chat model does not add those brackets automatically—ask the agent to include them in the visible reply (or set expectations in the agent system prompt) before using Read aloud. See Orpheus v1 English — Vocal Directions.
- Chat Window Preferences
- GT Chat
- Building Agents — agent Speech Settings
- Model Catalog — STT/TTS capability visibility
- Model Guides — provider-specific TTS usage (Orpheus vocal directions)
- Model Cards — official Groq and Gemini vendor documentation
- Settings — tenant appearance defaults
- Models — platform multimodal gates and default speech models