Platform · Voice agents

One agent, every customer conversation.

Define an agent once. Vocily AI applies the same role, voice, knowledge, and workflow rules across chat, browser voice, and outbound calls.

Problem · Solution

The problem today

Off-the-shelf voice AI gives you a 'helpful assistant' that doesn't know your business — wrong tone, wrong language, no rules about when to escalate. Production conversations need scope: what the agent is hired to do, which languages it can switch between, what it can quote, and exactly when it must hand off to a human. Otherwise the agent improvises in places it should be quiet and goes quiet where it should be confident.

How Vocily AI handles it

  • Job-scoped role

    Define what the agent is hired to do, the tone it should take, and the topics it must refuse — its prompt enforces the boundary, not your training docs.

  • Language and voice per agent

    Indian English, Hindi, Hinglish auto-switch, Bengali, Gujarati, Kannada, Malayalam, Marathi, Tamil, Telugu and more — picked per agent, switched live by the caller.

  • Knowledge boundary

    An agent answers from the knowledge base attached to it. Other agents' KBs aren't visible — no cross-talk between billing and sales.

  • Workflow rules, not vibes

    When to use the KB, when to call your APIs, when to book a meeting, when to transfer — declared once, applied to every call.

  • Escalation as a first-class rule

    Set the conditions that must trigger a human handoff. The agent doesn't decide; the rule does.

  • One agent, every channel

    The same role, voice, and rules run on chat, browser voice, and outbound calls — no parallel configurations to drift out of sync.

What's in it

What you configure on a Vocily AI agent.

Every Vocily AI agent is built from a small set of explicit fields. No hidden behaviour — what you set is what the agent does.

Identity & role

The agent's job and how it should sound saying it.

Name
Display name shown in the dashboard and on transcripts.
Role prompt
Up to 64 KB. Built-in {{caller_no}} for inbound; custom runtime variables injectable per call via API. Section editing for purpose / response style / workflow.
Greeting
First line the agent says when a call connects.
Conversation initiator
Bot greets first, or wait for caller to speak first — per agent.
Voice Style Preamble
Auto-applied 'speak naturally, no markdown, spelled-out numbers' guardrail prefixed to the prompt.
Welcome
Widget-side intro for chat and browser-voice sessions.

Language & voice

Speech-in, speech-out, and the languages the agent can switch between.

Languages
Indian English · Hindi · Hinglish auto · Bengali · Gujarati · Kannada · Malayalam · Marathi · Odia · Punjabi · Tamil · Telugu · Urdu · Assamese (17+)
Voice
Provider-routed TTS voice — Cartesia, ElevenLabs, Sarvam, or Smallest. Per-language voice mapping inside the same agent.
STT
Provider-routed transcription with word-level alignment events; keyword boosting per agent for brand names and SKUs.
Pronunciation dictionary
Custom word → spoken-form mapping, provider-agnostic.
Audio ambience
Optional low-volume background sound (office chatter, keyboard typing, none) to mask LLM latency.

Knowledge & tools

What the agent can quote, look up, or act on.

Knowledge base
One or more KBs attached per agent — websites, documents, FAQs.
Toolkit
Explicit list of API tools the agent is allowed to invoke.
Scheduling
Calendar tool for in-call booking and reschedule.
Transfer
Named human destinations the agent can hand off to.

Listening behaviour

How the agent decides when to speak, when to stop, and what to ignore.

Interruption sensitivity
Tunable 0.0–1.0. Filler 'umm' / 'hmm' is ignored; only real turn-starts cut the agent off.
Turn detection
Configurable response eagerness — patient ↔ eager — for how fast the agent jumps in after a caller finishes.
Noise handling
Off · Standard · High · Background-voice strict guard (actively rejects voices that aren't the primary caller).
Silence handling
Auto-end on prolonged silence (default 25s). Optional pre-timeout 'Are you still there?' check-in at a custom delay.

Escalation & handoff

When the agent must stop talking and hand off.

Triggers
Topic match, sentiment threshold, explicit caller request, KB-empty fallback.
Destination
Phone number, internal team, or queue — per trigger. Business-hours-aware.
Warm vs cold
Warm transfer briefs the human first; cold transfer dials straight through.
Handoff payload
Conversation summary, language used, caller fields collected.
After handoff
Recording stops; pre-handoff transcript persists on the execution.

Lifecycle & versioning

Edit safely. Test before live traffic. Roll back in one click.

Drafts
Edits live in a draft until you publish. Live channels keep running the published version.
Test surfaces
Dedicated test call, test chat, test widget — always run your latest saved draft.
Publish
One click to promote draft → live. All channels switch at once.
Rollback
Revert to any prior version in one click. KB content stays shared; only the agent config rolls back.
Starter templates
Three production-ready templates — customer support, lead qualifier, appointment booker — pre-configured prompts, voice, tools, KB seeds.
Consent placement
Before greeting, after greeting, or skipped. Configurable per agent.

Same agent, every channel

Consistency where customers expect it.

An agent's persona, knowledge access, and escalation rules should not change just because the conversation moved from your website to a phone call. Vocily AI keeps them aligned.

Chat widget
Browser voice
Outbound calls

One agent, every channel

Same voice. Same knowledge. Same rules.

Whichever channel a customer reaches you on, the agent picks up where the conversation left off.

Active channels

Chat widget
Browser voice
Outbound calls

Actions the agent can take

Book a meetingLook up the callerTransfer to a teammateSend a follow-up SMS

Configured for real conversations

What you set up, the agent does.

A Vocily AI agent reads from a clear configuration — its job, the languages it speaks, the knowledge it can reference, and the rules for when it should escalate. There is no hidden behavior.

What the agent is hired to do
Languages it can switch between
Knowledge it can quote
When it must transfer to a human
Channels it works on
Daytime + after-hours behavior

Insurance renewals agent

What this agent does

Picks up renewal-due policyholders in Hindi or English.

Live
JobConfirm intent to renew, answer premium questions, book a callback if needed.
LanguagesHindi · English · Hinglish (auto-detects)
Knowledge it can useYour policy docs, FAQ, current premium table
Must transfer whenCaller asks for a human, complaint, or compliance topic
After-hoursBooks a callback inside the call instead of leaving a voicemail
OutboundChatBrowser voice

Common questions

What teams ask before they switch.

One. The same agent definition serves chat, browser voice, and outbound calls. Its role, persona, KB access, and escalation rules are identical across channels — no parallel configurations to maintain.