Your agents
Models
Used for widget voice calls (OpenAI Realtime — speech recognition, the model and the voice in one connection). Independent of the chat model above.
How long after the page loads before the agent greets the visitor. 0 = immediately. Applies to the chat widget; voice calls greet as soon as the call connects.
What the agent says at the start of every conversation, on calls and in chat. There is one greeting; the Chat tab used to show a second box for the same value.
Tell the agent who it is, how it should behave, and what it knows
FocusedCreative
Sent to the agent as the exact sentence to say. Needs ticketing on — otherwise it promises a route the widget does not have.
Never promise a named person or a time — you do not control who is available.
Sets both the wording and the ceiling: concise 160 tokens, balanced 300, detailed 700.
A per-agent cap within the setting above — it can lower the ceiling, never raise it.
Knowledge base
Paste any content directly — FAQs, product info, scripts
0 characters
Free — no storage cost
File upload is not available yet
Open the file, copy the text, and paste it into Add text. That is stored and used exactly the same way.
Auto-fetch pricing
Manual fetch: Free
Daily auto-fetch: ~$0.001/day
Weekly auto-fetch: ~$0.007/month
Monthly auto-fetch: ~$0.001/month
Cost covers server processing to fetch and index page content. Set frequency per URL after fetching.
Knowledge base cost: $0.00/month (text only — free)
Cost breakdown for this agent
STT (Speech-to-Text)Deepgram Nova-3 · ~$0.005/min
TTS (Text-to-Speech)OpenAI TTS · ~$0.011/min
AI ModelGPT-4o Mini · ~$0.001/min
Platform services hosting, routing & platform~$0.016/min
Base cost~$0.033/min
Your plan price~$0.055/min
Agent voice
All voices support 40+ languages. Select your language and the voice will automatically respond in it.
Slow (0.5x)Fast (2.0x)
Adjust how fast or slow the voice speaks
Tune if the agent gets interrupted by background noise, or is slow to notice the caller.
Filters noise before speech detection. Far field suits support calls.
Sensitive (0.00)Strict (1.00)
How loud speech must be to count. Raise it if noise interrupts the agent.
How long the caller must pause before the agent replies. Raise so a brief noise does not end their turn.
Audio kept from just before speech starts, so the first word is not clipped.
Assign a Twilio phone number to this agent for inbound/outbound calls
Speech recognition
Recommended
Deepgram Nova-3
$0.0049/min
Latest, most accurate for voice agents
Pro
Deepgram Flux
$0.0059/min
Built-in turn detection, lowest latency
Deepgram Nova-2
$0.0043/min
Fast and reliable, previous generation
OpenAI Whisper
$0.006/min
Great for accented speech
Google Speech
$0.016/min
Best for non-English languages
Accept allHigh confidence only
Transcripts below this confidence score are ignored
Help the AI recognise these words accurately (product names, technical terms). Comma-separated.
How the agent listens
Instant2 seconds
How long the agent waits after you finish speaking before replying
Instant10 words
Customer must speak this many words before the agent stops talking
0.1s1s
Agent stops speaking after customer talks for this long
0s5s
Time before the agent speaks again after being interrupted
0.5s10s
How long to wait after speech ends before agent responds
Voicemail
Escalation to human
📊 Widget Activity
Active visitors
0
Chats today
0
Callbacks today
0
Bookings today
0
🔧 Installation
Paste this snippet anywhere on your website. Works with all platforms including WPCode, Google Tag Manager, and any tag manager. The widget auto-detects logged-in users, fetches all settings from the server, and just works.
Where to paste
WordPress (WPCode): WPCode → Add Snippet → HTML Snippet → Footer
WordPress (theme): Appearance → Theme Editor → header.php → before </head>
Shopify: Online Store → Themes → Edit code → theme.liquid → before </body>
Google Tag Manager: Custom HTML tag → Fire on All Pages
Wix: Settings → Custom Code → Add code → Body end
Squarespace: Settings → Advanced → Code Injection → Footer
Webflow: Project Settings → Custom Code → Footer Code
React/Next.js: Add to _document.js or layout.tsx <head>
🎨 Appearance
Choose the launcher button shape and icon. Colour is controlled by the colour picker below.
Shown in the widget header. Leave blank for agent name.
⚙ Features & Behaviour
0 = disabled. Widget auto-opens after this many seconds.
Comma-separated URL paths where the widget should NOT appear.
Shown when this agent is switched off — the visitor sees this instead of the chat. It is not a schedule: it appears whenever the agent is off, not outside any set hours, so avoid wording like “back at 9am”. If ticketing is on, the widget still offers the ticket form so visitors can reach you. Leave blank and the widget does not appear at all.
The widget auto-detects logged-in users by scanning cookies, localStorage, meta tags, WordPress DOM, and common window objects.
Auto-detect works on most sites. Use meta/window if you need explicit control.
🤖 Feature Toggles
Display widget on mobile devices
AI suggests contextual reply buttons
Chime when a reply arrives in the visitor's widget
Display "Powered by Vokeva" in widget
Take over chats from Live page
Voice calls
Controlled by the agent Type setting on the Model tab.
Abuse limits
agent_id is public in the embed snippet on every site it appears on.
0 = unlimited. Bounds spend even if someone cycles IPs.
Blank = any site. Browser-enforced only — it stops the snippet being copied onto an unapproved site, not a determined attacker. The daily limits are the real bound.
Visitors request phone callbacks
Show a ticket submission tab in the chat widget
👤
Pre-chat Form
Collect visitor info before the first message. Visitors who provide their name will be identified in the Live page and Chat Logs.
Built-in actions
Let callers press keypad digits during a conversation (e.g. press 1 for sales). Configure in Call Behavior tab.
Custom webhooks
Webhooks notify your systems when events happen
Humanoid AI layers — phone & interview-room calls
✓Layer 3 — Natural speech patterns
✓Layer 6 — Prosodic mirroring
✓Layer 8 — Session topic memory
✓Layer 2 — Adaptive follow-ups · interview agents
✓Layer 4 — Emotional intelligence · interview agents
✓Layer 5 — Ultra-low latency TTS · requires a Cartesia voice
○Layer 7 — Backchanneling · disabled; it cut across the caller
These run in the server-side voice pipeline, which carries phone calls and interview-room sessions. Widget voice calls take a different path — the browser connects straight to OpenAI Realtime, where one model does the listening, thinking and speaking in a single connection. It handles turn-taking, interruption and prosody natively, so there is no separate pipeline for these layers to sit in. Nothing is missing from widget calls; they are built differently.
What to track from every conversation
Call summary
Success tracking
Examples: "Customer issue was resolved" / "Appointment was booked" / "Customer expressed satisfaction"
Success rate
Scored by post-call analysis on phone & interview-room calls. See the score per call in Call Logs.
Alerts
Call limits
10s120s
Hang up if nobody speaks for this long
1 min60 min
Automatically end the call after this time
Privacy
Keypad input (IVR)
Silence filler messages
Messages the agent says when the caller goes silent
How many times to prompt before ending the call
5s60s
Seconds of silence before saying a filler message
Call ending
What the agent says just before ending a call
Voicemail detection method
Presets
Optimised for long-form agents: long silence tolerance, 20-min calls, patient turn-taking
Call logs
| Call ID | Assistant | Type | Status | Sentiment | Score | Start Time | Duration | Minutes |
|---|
Test Chat
×
Chat History
Loading chats...
Select a conversation to view
Online now
0
Chatting
0
Idle
0
Exit intent
0
Loading visitors...
Select a visitor to see details
Unique visitors
0
Activity
0
heartbeats, not visits
Identified
0
Engaged
0
stayed beyond one check-in
Activity and Engaged are derived from heartbeats, not visits — the widget checks in
at least every 5 minutes while a visitor is on the page. Per-visit counting is not
recorded, so these cannot be converted into visit counts, historically or going forward.
Select an agent to view visitor profiles.
Phone Numbers
Setup guide
1. Get your credentials at twilio.com/console → Account Info
2. Buy a number at twilio.com/console → Phone Numbers → ~$1.15/month
3. No Twilio account? Sign up free at twilio.com
Your credentials are stored securely and used only for your agents' calls.
API keys
API keys let external systems interact with your agents.
Coming soon
Billing
Current plan
Free
Voice minutes
0 / 20
Chat messages
0 / 100
Tickets
0 / 100 (free)
Active agents
0
Upgrade your plan
Starter
$29/mo
3 agents
500 minutes/month
Phone support
500 minutes/month
Phone support
Pro
$99/mo
10 agents
2,000 minutes/month
Priority support
Analytics
2,000 minutes/month
Priority support
Analytics
Agency
$299/mo
Unlimited agents
Unlimited minutes
White-label
Custom domain
Unlimited minutes
White-label
Custom domain
White Label Settings
Brand your platform
Saved to your agency profile. Serving the dashboard on your own domain is not available yet — contact us before setting up DNS.
Sub-accounts (clients)
A/B Tests
Support Tickets
| Ticket # | From | Agent | Reason | Status | Priority | Created |
|---|