​

Voice Agent channel toggles walkthrough

​

1. Where the surface lives

• Agents are built in the AI Agents hub. You create, prompt, and activate agents under AI Agents. This channel page never edits an agent; it binds one agent to the web-call channel and sets the channel-scoped behavior around it. The picker lists active agents only, because a call can only be answered by an active agent.
• The binding is channel-scoped. The same agent can serve other channels, each with its own settings. Everything you set here applies to the web-call widget only.

​

2. The toggles, explained

​

Agent and status

• Enable the voice agent widget is the master switch. While it is off, an installed snippet places no calls.
• AI voice agent is the active agent a visitor talks to. If you have no active agents yet, the page says so and points you to AI Agents; create and activate one there first.

​

Appearance

• Brand color and launcher position set where the floating button sits and which of your brand colors it uses. These are the same two controls as the web chat widget’s branding, stored against this widget’s own config.
• Theme styles the button light or dark to match your site.
• Button label is the text on the floating button, up to 60 characters. The default is “Talk to AI”.

​

AI disclosure and consent

• AI disclosure text (up to 500 characters) is the notice shown to the visitor before a call connects, so they know they are speaking with an AI.
• Require explicit consent before the call starts is on by default. While it is on, the visitor must accept your consent copy (up to 1000 characters) before the call connects.

​

Business hours

​

3. Who can change what

​

4. Verify via the API

curl -s https://orbit.devotel.io/api/v1/settings/voice-agent-widget \
  -H "Authorization: Bearer $ORBIT_API_KEY"

curl -s -X PUT https://orbit.devotel.io/api/v1/settings/voice-agent-widget \
  -H "Authorization: Bearer $ORBIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "enabled": true,
    "agent_id": "agt_your_active_agent",
    "primary_color": "#6366f1",
    "position": "bottom-right",
    "theme": "light",
    "button_label": "Talk to AI",
    "ai_disclosure_text": "You are about to speak with an AI assistant.",
    "consent_required": true,
    "consent_text": "By continuing you agree to talk with an AI. A human can join at any time."
  }'

​

5. Common failure paths

• No agent answers. The picker was left on “Select an agent…”, or the bound agent was archived after binding. Only active agents are listed and can answer; rebind under Configure and save.
• The widget does nothing on the site. The enable switch is off, or the snippet from the Install tab was never pasted (or was pasted on a different domain than the one you test). Re-copy the snippet from the Install tab and confirm the enable switch is on.
• Missing or stale disclosure. The channel copy was never written, or the workspace posture changed and the channel copy no longer matches it. Update the posture under Settings → Compliance → AI Disclosure, then rewrite this channel’s disclosure and consent copy, and re-run the verification steps in AI disclosure pre-publish setup & evidence.
• Consent required but no consent copy. If you keep Require explicit consent on (recommended wherever you need an affirmative record), write the consent copy visitors accept; an empty consent line gives visitors nothing to agree to.
• Hours behave oddly. The windows are evaluated in the workspace timezone, not the visitor’s. A “9 to 5” schedule means 9 to 5 in your workspace timezone.
• The Save button stays disabled. The page tracks edits against the saved config and enables Save only when something actually differs. It is not stuck; change a field and Save becomes enabled.

​

6. Where it fits

• AI disclosure pre-publish setup & evidence: the workspace-level disclosure posture this channel’s copy implements, plus the evidence trail for audits.
• Channel settings consoles: the organization-level settings consoles this page sits alongside.
• AI Agents in the dashboard: where the agent you bind here is built, prompted, and activated.