Voice agents - Bolna Docs

A Bolna voice agent listens to a caller, decides what to say, and answers in a natural voice — on inbound calls, outbound calls, batch campaigns, or in the browser. Every agent follows the same lifecycle: build it, deploy it to a phone number or app, then monitor what happens on each call. When it works in testing, go live with the launch checklist.

Build

Set the prompt, LLM, voice, languages, tools and knowledge base in Agent Studio, or design a multi-step flow with graph agents.

Deploy

Buy or connect phone numbers, place outbound, batch and workflow calls, receive inbound calls, or bring your own SIP trunk.

Monitor

Review call history, receive webhooks, pull executions over the API, and extract structured data from every call.

Go live

Test before calling real customers, control cost per call, handle failures at scale, and troubleshoot common problems.

Featured

Graph agents

Part of Build. Design a multi-step conversation as typed nodes and routing rules instead of one long prompt.

Workflows (beta)

Part of Deploy. Chain calls, WhatsApp messages, conditions and retries into one sequence per contact, and run it over contact lists as campaigns.

Not sure which one you need? See graph agents vs workflows.

Build your first agent

Quickstart

Describe your use case and get a call-ready agent from Agent Studio in 5 to 8 minutes.

API quickstart

Place an outbound call and fetch the result over the API.

Inbound quickstart

Link a phone number to an agent and answer incoming calls.

Browse templates

Start from a ready-made agent for support, sales, reminders, bookings and more.

Platform capabilities

Build

Capability	What it does
Prompting	Write the agent’s instructions and personalize them per call with dynamic variables
LLM	Pick the model that reasons for your agent, or bring your own
Voices	Choose the voice for each language, or clone and import your own
Languages & transcription	Switch languages mid-call, write prompts in other languages, and improve recognition with keywords
Engine	Tune response latency and how the agent handles interruptions
Knowledge base	Ground answers in your PDFs and URLs
Tools	Transfer calls, book calendar slots, and call your own APIs mid-conversation
Graph agents	Design multi-step conversations with typed nodes and routing
Templates	Start from a ready-made agent for your industry

Deploy

Capability	What it does
Phone numbers	Buy numbers from Bolna, including 140 & 160-series numbers in India
Outbound calls	Call one person, or thousands with batch calls, auto-retry and calling guardrails
Workflows (beta)	Chain calls, conditions, retries and WhatsApp messages into one sequence per contact, run them as campaigns, and get a webhook when each finishes
Inbound calls	Answer calls, identify callers and route them through an IVR
Telephony	Connect Twilio, Plivo, Exotel or Vobiz, or bring your own SIP trunk
Web calls	Talk to your agent from the browser with the Web Call SDK

Monitor

Capability	What it does
Call history	Listen to recordings and read transcripts, costs and latency for every call
Call analytics	See how many calls complete and become conversations, and the best time to call
Latency	Understand and measure where response time goes
Webhooks	Receive call updates on your server as they happen
Extractions	Pull structured data such as outcomes, dates and intents out of each conversation
Call statuses	Understand every state a call moves through and why it ended

Go live

Capability	What it does
Launch checklist	Verify telephony, compliance, quality, observability, concurrency and cost before real calls
Test before calling customers	Check your agent end to end before it reaches real people
Control cost per call	Reduce spend without hurting quality
Failures & retries at scale	Classify failures, set retry policy and monitor large campaigns
Troubleshooting	Fix dropped calls, audio problems, interruptions, failed transfers and missing webhooks

How a call works

A standard agent chains three services: a transcriber turns the caller’s audio into text, an LLM decides what to say, and a synthesizer turns the reply back into audio. A speech-to-speech agent replaces all three with a single realtime model. See how a call flows through Bolna for the full lifecycle, from dialing to the finished execution record.

Build

Deploy

Monitor

Go live

​Featured

Graph agents

Workflows (beta)

​Build your first agent

Quickstart

API quickstart

Inbound quickstart

Browse templates

​Platform capabilities

​Build

​Deploy

​Monitor

​Go live

​How a call works

Featured

Build your first agent

Platform capabilities

Build

Deploy

Monitor

Go live

How a call works