Quickstart

Boltun makes real phone calls for you. You describe what you sell, it writes the agent, and it works through your list — talking, handling objections, and handing you a transcript and an outcome for every call. Here is the whole thing, end to end.

How it works

Four things make up everything you will touch in the app.

When a campaign is running, the dialer picks the next lead that is due, checks it is a sensible hour where that person lives, places the call, and writes the result back. It keeps several calls going at once and stops on its own when the list is finished, when you pause it, or when your balance runs out.

Around the app

The left sidebar is the whole app.

Open a campaign and you get five tabs across the top — Design, Test calls, Leads, Calls and Settings — plus Start dialer, Pause and Reset in the top-right corner.

Design a pitch

The Design tab is a chat with Robert, the agent designer. Tell it who you are calling and what you are offering, in plain words. It drafts the opener, the answers to the objections you will actually hear, and the close — then turns that into the instructions your caller follows. You never have to write those instructions yourself, but you can read and edit them at any time.

Robert keeps the context of the campaign, so you can come back later — “be warmer at the start”, “stop offering a discount” — and it reworks the pitch instead of starting over.

Robert finishes his thought. If you refresh the page, switch tabs, or lose your connection while he is mid-reply, he carries on writing in the background — the finished answer is waiting when you come back, and it is saved to the chat either way. You never have to re-ask just to recover a half-written response.

He stays in step with your settings. Change something on the campaign’s Settings tab mid-conversation and a Settings updated card appears in the chat, listing only the fields whose value actually moved. Edit five things in a row and you still get one card, not five — it keeps absorbing your changes until Robert has been told about them. That same note reaches him, so his next suggestion accounts for it instead of talking about the old setup. And when he wants to change how your agent is configured, that change is put to you to accept — nothing about your agent is rewritten silently (see Approvals). The post-call side of things is handled by a second assistant, Richard — see After the call.

He can look at your leads. Ask “how is this campaign doing?” and Robert reads the same headline you see under the campaign title — total contacts, how many are called, how many are left, anyone on a call right now, and the count for every outcome. Ask about the people themselves and he reads the same Leads table you see — name, number, timezone, language, status and attempts — so “how many are still pending?” or “what happened with the callbacks?” get answered from your actual list instead of a guess. He can narrow it to one status, and he can open a single person by phone number to see their notes, any callback that is booked, and a summary of every call they have had — then pull up the full transcript of any one of those calls if it matters. He only ever reads: nothing about a lead is changed from the Design chat.

Test calls

Before you spend anything on a real list, hear it. The Test calls tab lets you talk to your own agent two ways:

The transcript appears as you speak, and each turn shows how long the agent took to hear you, think, and answer — so if a reply feels slow, you can see which part caused it. You can also type instead of talking, which is quicker when you are only checking wording.

Leads & import

Add people one at a time with Add lead, or bring a list in with Import CSV. On import you are shown how your columns were matched before anything is saved, and any rows that could not be read are listed back to you rather than dropped silently.

# a CSV that imports cleanly
phone,first_name,surname,timezone,language,notes
+14845550107,Jordan,Ellis,America/New_York,English,renewed last spring
+442071838750,Priya,Shah,,,asked to be called after 6pm

Only the phone number is required — not even a name. A lead with no name simply shows as Unknown until you add one, so you can bring in a bare list of numbers and start dialing. Leave timezone out and Boltun works it out from the number itself — see Scheduling & timezones. Anything you put in Notes is given to the agent before it dials, so it can open with something specific to that person.

The language column pins a person to a language of their own; leave it blank and they follow the campaign’s default. See Languages for how the campaign default, a person’s own language, and Auto fit together.

Every field except the phone number is now labelled optional wherever you meet it — on the Add lead form and in the import column guide — so you can see at a glance what you can safely leave blank.

The global Leads page in the sidebar is your whole address book, across every campaign. Search by name, phone or company, and select rows to add them to a campaign. Click any lead to open their card: details, call history, and the notes the agent will see. You can also change a lead for one campaign only; when you do, a Revert to master link appears so you can put the original details back. Where a call stands — its status — lives inside each campaign, not on this global page, because the same person can be at a different stage in two campaigns at once.

Statuses live in the campaign. A campaign’s Leads tab shows the same people with the same notes as the global page — click a name there and you get the identical card — but it adds one thing the global page cannot: every contact carries a coloured status showing exactly where it got to:

You do not need to refresh to watch a run. The contact list keeps itself current while a campaign dials, so statuses move from Pending to Calling to whatever they end up as in front of you. The campaign header above it counts the same thing in words — how many contacts there are, how many have been called, how many are left, and how each of the finished ones turned out. It follows every change you make too: reset the campaign and the line re-reads the run from scratch rather than leaving yesterday’s tally on screen.

Hover any status to read what it means and what it means for that particular contact. The tooltip opens with a short plain-language explanation of the status, then adds this contact’s own detail — a Callback tells you the time it is booked for, a finished call tells you when it ran and how many attempts it took, and a Pending contact that has already been tried tells you which attempt is next — followed by any note the agent left. To put a contact back in the queue, use ↻ Reset status in its row menu, which sets it back to Pending for another attempt.

Languages

Boltun can call people in their own language. You set it once for a whole campaign, override it for individual people where you need to, or let the agent work it out on its own — and when more than one of those is in play, the most specific one wins.

Set a campaign default. On the campaign’s Design tab, under Agent configuration, choose the language your agent speaks. Two ways to go:

Set it per person. Give a lead its own language — in the Add lead form, on their card, or the language column of your CSV — and Boltun speaks to that one person in it, whatever the campaign is set to. Leave it blank and they simply follow the campaign default.

Or just for one campaign. The same person can sit in two campaigns that run in different languages. Open the lead from inside a campaign and set their language there and it applies to that campaign only — a Revert to master link puts it back to their usual one.

Which one wins. If a person has a language set more than one way, Boltun takes the most specific:

Every call records the language it actually ran in, shown on the call in the Calls tab — so you can confirm Auto matched the right one.

In-call tools

Talking is most of the job, but a call sometimes needs an action — hanging up, getting past a menu, changing language. Your agent has a small set of tools it can reach for on its own, at the right moment, without you steering it.

You decide which tools are available on the campaign’s Settings tab, under Tools — or for every campaign at once under Settings › Agent tools. Each one is listed with a one-line summary of what it does and a switch to turn it on or off; hover the summary to read the fuller detail. The descriptions are there to read, not to edit.

What the agent remembers. If you have called this person before, the agent starts the call already knowing a short recap of each earlier conversation — not the transcripts. That is deliberate: a handful of whole transcripts would crowd out your actual pitch and slow every reply. So it opens with “we spoke in March and you asked us to try again after the quarter”, and only reaches for the full transcript with the tool above if the detail actually matters. Nothing is carried between different people — each call sees only that contact’s own history.

Calls & transcripts

The Calls tab is the record of everything that happened. Calls in progress appear at the top with a live badge and their transcript streaming in; finished calls keep the full transcript, the recording, and how long they lasted. The transcript ends with a Call ended by line telling you who hung up — the agent, once it was done, or the person on the other end — so a call that was cut off short is easy to tell apart from one that finished naturally.

Every completed call is then read back by Richard, the post-call analyst, who writes a short summary and marks the outcome — interested, not interested, wrong person, no answer, and so on — so you can skim a hundred calls without listening to any of them. While he is still working on a just-finished call, its summary shows a loading placeholder that fills in on its own the moment he is done. There is more he can do than describe the call — see After the call.

Each row shows which way the call went: ↗ OUT for one you made, ↙ IN for one that came to you. See When they call back.

After the call: Richard

Boltun has two assistants, and it is worth keeping them straight. Robert is the one you chat with on the Design tab before you dial — he writes the pitch. Richard works after the fact: he reads every finished call, decides what it means, and writes the summary and outcome you see on the Calls tab.

Richard’s read-back comes back as a tidy, structured result rather than a wall of text — the outcome, a short summary, and anything he did about it, laid out the same way every time. While he is still analysing a call the summary shows a brief loading placeholder, then swaps itself for the real thing without you refreshing.

He can act, not just describe. Where it is warranted, Richard updates your data for you off the back of a call:

Everything he does is on the record. Anything Richard changes shows up twice: as a small chip right in the transcript at the moment he did it, and as a plain-English Actions taken list under his analysis — “Moved the contact from Interested to Booked” — so there is never a change you cannot trace back to the call that caused it.

How Richard reads and reacts to calls is set by your workspace administrator, the same way Robert’s own instructions are — so the read-back stays consistent across your whole team. You do not need to configure anything for it to run.

When they call back

People miss calls, and a good number of them ring back. Boltun answers. It looks up the last call you made to that person from the number they dialled, and picks the conversation up with the same agent — same instructions, same voice, same engine, same settings as the call they missed.

The agent knows the person rang you, not the other way round, so it opens by greeting them and asking how it can help rather than launching into the cold-call pitch it would use on an outbound call. It is also told why they are probably calling: someone returning a call that never connected is greeted differently from someone who spoke to you yesterday and has a follow-up question — it knows which of the two it is before it says hello.

The call lands in the same campaign as the one that reached out, marked ↙ IN instead of ↗ OUT, with the transcript, recording and summary you get from any other call. Inbound calls never count toward auto-refinement — that measures how your pitch performs when you make the call.

Someone you have never called. Every number you own gets its own campaign, named INBOUND +1…, created for you the moment the number is assigned. That is what answers strangers on that line. Give each one its own instructions and a caller who found your Berlin number gets a different greeting from one who rang your New York number. These campaigns are tied to their line, so they cannot be renamed, deleted, or pointed at a different number.

An inbound campaign has no list to import — its leads walk in. Start it the same way you start any campaign; it will not ask you for contacts, because there are none to dial. What it does need is instructions: a line with an empty prompt is refused when you press Start, rather than leaving callers to hear the carrier say the number is unavailable. Every stranger who rings is then added to that campaign’s Leads as they call, filed under Unknown until the agent learns a name, and from then on they are an ordinary lead: notes, language, status, call history, the lot.

Answering is off for a line the moment any one of three switches is off: the workspace switch your administrator controls, the Answer inbound calls switch on your Settings page, and the one on each campaign’s Settings tab. Both of those switches stay greyed out until you have a number of your own: the free shared line is used by other accounts, so a call to it cannot be matched to your agent no matter how the switches are set. When a call is turned away the reason is written into the call log, so you can see it happened and why.

One requirement: the number has to be yours alone. A number shared between accounts is tagged SHARED under My numbers and cannot answer, because there is no way to tell whose agent should pick up. You can buy a dedicated caller ID of your own from Settings, or ask your administrator, if the switch is greyed out.

Scheduling & timezones

Set the hours you are willing to call in and Boltun will not dial outside them. You choose a start time, an optional end time (leave it empty to run until midnight), and which days of the week are allowed. Turn scheduling off entirely and it calls whenever the dialer is running.

Recurring days, with exceptions. Weekdays are the backbone — pick the days a campaign is allowed to run, Monday to Friday by default. On top of that, a small month calendar handles the odd day out: tap a date to skip it (a holiday you don’t want calls going out on) or to add it (a one-off push on a day the pattern would normally sit out). And if you would rather space a campaign out, an optional every-N-days setting runs it once every few days instead of on a weekly rhythm.

The part that matters is whose clock those hours follow:

How the contact’s timezone is worked out. Boltun takes the first answer it can get, in this order:

You can always see and change what was detected: open the lead and the timezone field shows what Boltun is using. All 448 standard timezones are available, and the field is searchable.

Calling regions

A phone conversation only feels natural if the reply comes back fast. To keep that true wherever you are calling, Boltun runs your agent from four places around the world and uses the one closest to the person being dialled:

The choice is automatic, and it uses the same timezone Boltun worked out for scheduling — so a lead in New York is handled from Ashburn and one in Berlin from Frankfurt. The audio takes a shorter path and the pause before each reply gets noticeably smaller. There is nothing to set up: import a list from anywhere and each call routes itself. If a number cannot be placed at all, the call is handled from Frankfurt.

Auto-refinement

Your first pitch is rarely your best one. Boltun can pause a campaign automatically after a set number of completed calls, read those transcripts, and come back with concrete changes — the opener people hang up on, the objection that keeps landing, the phrasing that works.

Set it on the campaign’s Settings tab: how many calls to review after (10 by default, 0 turns it off). Once the changes are accepted, the campaign carries on where it left off.

Approvals

If you would rather nothing changed behind your back, turn on approvals. Any change to how your agent talks then waits for you: a badge appears on the campaign, you see exactly what would change — the old wording next to the new — and nothing takes effect until you approve it. Approve and the campaign resumes on its own; reject and it keeps the pitch you already had.

Speaking style

Pick the voice on the campaign’s Settings tab — you can hear every one of them first on the Voices page, filtered by language, gender and character.

The same Settings tab is also where you choose the engine behind the voice — the technology that does the talking. There are several, and they trade off naturalness against cost, so it is worth trying more than one. Alongside the established options you now have three newer engines to pick from: Grok 2.0, OpenAI Realtime Mini — a cheaper tier of the OpenAI voice — and Inworld, the most economical of the lot. If you are watching the budget, Inworld or the Mini tier stretches it furthest; the per-minute rate for every engine is on the Pricing page.

You do not have to leave the page to compare them. Open the engine dropdown and every option carries its own figures on the right — a quality score out of ten, a speed figure measured from recent real calls, and the price per minute you would actually be charged. The speed column names what it is timing rather than averaging everything together: TTFT on a brain, the wait for its first word, and TTFB everywhere else, the wait for the first sound the caller actually hears. A whole engine reports TTFB only — it speaks straight out, with no separate thinking step to time. Click any of those three column headings to sort by it, and click again to reverse; the list opens sorted by quality, best first. Every speed figure you see is measured — there are no published estimates in that column. An engine, or in a custom stack an individual part, shows a dash until it has enough recent calls of its own to report a median honestly, and shows its real one from then on. Nothing is ever borrowed: a smaller sibling of a bigger model reports its own speed or none at all, so two figures side by side are always two measurements.

How hard it thinks. Next to the engine sits a reasoning level — how much the agent turns a reply over before it speaks. Nudge it up and answers come back more considered but a little slower; drop it and they are quicker and more to the point. Like the engine list, every level now carries its own measured speed on the right, timed from real calls, so you can see what a step up actually costs you in pause before you choose it — and a level shows a dash until enough calls have run at it to report an honest figure. On the all-in-one engines — Gemini, the two OpenAI voices and Inworld — each step up the scale adds a little to the per-minute rate, because the engine does more thinking on every turn — so you only ever pay more when you ask it to think harder. In a custom stack the level moves the speed alone, never the price. Not every engine takes the control: Grok runs at a single fixed depth and simply ignores it.

There is also a build-your-own option. Pick Custom Cascade Stack and you compose the pipeline yourself in three parts — the ear that turns speech into text, the brain that decides what to say, and the voice that speaks it — choosing each one separately and mixing makers freely. Every combination works; parts from the same maker usually share a region and shave a little latency, while crossing makers is fully supported. The price just follows the parts you pick and is shown live as you choose them, with the per-part rates on the Pricing page. The voice list follows the third part — whichever engine you set as the one that speaks — so the voices offered are always the ones your stack can actually say. Swap that part for another maker and a voice the new one cannot speak is cleared, ready for you to pick again — and if you leave it cleared, the stack simply uses that engine’s own default voice, so a campaign can never go out mute. A few of the newer parts are still being tuned and wear a small BETA tag in the picker, so you can tell them from the settled ones — and that tag comes and goes with your stack rather than being fixed to a part, because the caution behind it does not always apply. Choosing Inworld TTS-2 as the voice, for instance, sends the thinking through Inworld too, on their capacity rather than ours, and the Groq brains drop their BETA tag for as long as that is the case. The voice list carries two Cartesia Sonics, and the newer Sonic 3.6 has come out of beta and wears no tag at all. It speaks the same voices as 3.5 and switches language mid-call just the same, so auditioning it costs you nothing but a listen — it simply answers the speed control more faintly. Leave that part alone and you stay on Sonic 3.5, which is what every campaign speaks through today. One part, Inworld STT, is marked INWORLD ONLY: it works solely beside Inworld TTS-2, and picking it selects that partner for you. And a custom stack is not for one campaign only — set one as your workspace default on the sidebar Settings page, the same as any engine, and every new campaign starts from it.

A beat before it speaks. You can have the agent wait a moment after the line connects before it opens, rather than talking the instant someone picks up — a short, human pause that keeps the call from feeling automated. Set it in milliseconds as a workspace default on the sidebar Settings page, or per campaign on its own Settings tab. It only ever delays the first thing the agent says, and it is dropped the instant the other person speaks — or types, on a test call — first, so it never holds up a reply to someone who jumps straight in.

Beyond which voice, you can direct how it speaks in a sentence or two: “warm and unhurried, smile in the voice”, or “brisk and factual, no small talk”. Voices differ in how much direction they take, so change it, run a test call, and listen — it is the fastest loop in the product.

Sign in & security

Create an account with your work email and a password of at least eight characters containing a letter and a number. Boltun emails you a six-digit code to confirm the address, then asks for your company and team size — and you are in.

Forgotten passwords are reset from the sign-in screen: enter your email, take the code from your inbox, choose a new one. Keep me signed in holds the session for a month instead of a day.

Your Account page — under your name, top-right — is where you change your name, email or password, and where you switch on two-factor sign-in. With it on, signing in asks for a fresh code from your inbox as well as your password. Changing your email needs a code sent to the new address, so a stray session cannot quietly move your account somewhere else.

Balance & billing

Boltun is prepaid: you hold a balance and calls draw it down by the minute while they are in progress. No invoice at the end of the month, no seat fee. The Usage page shows what you have left, roughly how many minutes that buys at your current rate, and where it went — broken down by voice engine, by campaign, by destination country and by AI agent. The rate card there lists every voice engine’s per-minute price, the per-token prices for Robert and Richard, and the per-minute telephony cost for every country Telnyx reaches — scroll that column for the full list. A country shows a price once a call to it has been reconciled against the carrier’s own record; until then it shows a dash rather than an estimate.

You top up your own balance on the Billing page — reach it from the Top up button on your balance or the Billing link in the sidebar. Every movement — top-ups, call spend, adjustments — is listed with a date and a reason, so the balance always adds up. Per-minute rates for each voice engine are on the Pricing page.

Adding funds. Pick one of the ready-made amounts — $10, $25, $50 or $100 — or type your own, anywhere from $5 to $500, and tick Save this method for automatic top-ups if you want to keep the card for later. Continue to secure checkout hands you to the payment provider’s own secure page to finish — nothing is charged until you confirm it there. The moment the payment clears you are brought straight back to Boltun and your new balance appears on its own, no refresh needed. If your workspace has not turned payments on you will not see these controls, and your administrator tops the account up for you instead.

How you can pay. Any Visa, Mastercard or American Express card works, and so do Link, Apple Pay and Google Pay where your device offers them — they appear on the secure page on their own, so on a phone you can pay with a fingerprint and never type a card number. Boltun never sees or stores the card itself; the payment provider holds it and hands back only the brand and last four digits, which is all Boltun ever shows.

Automatic top-up. So a campaign never stalls for want of credit, you can let Boltun refill the balance for you. You set two amounts and pick one saved card: a threshold to watch for and a top-up amount to add. It starts switched off, and nothing is ever charged automatically until you turn it on yourself.

Once it is on, this is the whole of when it charges — there is no other trigger:

Turning it off. Switch automatic top-up off with the same toggle on the Billing page whenever you like; it takes effect at once and no further automatic charge can happen. It also needs a saved card to run, so removing the card it uses stops it just as surely.

Removing a card. Every card or wallet you have used is listed on the Billing page with its brand and last four digits. Remove next to one forgets it for good, and you will be asked for payment details afresh next time you top up. If you remove the card automatic top-up relies on, automatic top-up simply pauses until you point it at another.

Settings

There are two layers. Settings in the sidebar holds the defaults every new campaign starts from — voice engine, calling hours, whose timezone to use, when to auto-refine. A campaign’s own Settings tab overrides them for that campaign alone; leave a field untouched and it keeps following the default, so changing it once changes it everywhere. An untouched field is not blank — it shows the value it is inheriting, greyed out, so you can see what the call will use without opening the other page. A campaign has its own Reset to defaults beside Save: it drops every override that campaign holds and puts it back on your defaults, leaving its prompt, schedule and numbers exactly as they are.

These defaults are yours to edit — change the voice, calling hours, timezone or auto-refine and every new campaign you create starts from your version. Made a mess? Reset to defaults at the bottom puts everything back to the workspace baseline in one click; it never touches your campaigns or numbers. A few workspace-wide controls stay with your administrator: how the two AI assistants behave — Robert the designer and Richard the post-call analyst — and the global call limit.

The same page holds My numbers — the caller IDs assigned to your account, which is what your leads see when the phone rings. Each line carries small chips showing what it can do — voice, SMS, HD voice and the like — the same capability tags you see when buying a number, so you can tell at a glance what a number supports. Every account starts with at least one free shared number so you can place calls straight away; because it is shared with other accounts it is tagged SHARED and, as above, cannot answer inbound. The Answer inbound calls switch below the list controls whether your numbers pick up when someone rings them; see When they call back.

Buy your own number. When you want a caller ID that is yours alone — a local presence in a particular city, or a line that can take inbound calls — Buy numbers in My numbers opens a search over what is available. Filter by region, type (local or toll-free) and the features you need — voice, SMS, MMS, HD voice, international SMS, local calling or emergency — and search the country box by name or code to reach any country Telnyx sells in. Pick the ones you want one at a time, or tick Select all to take every number on the list, and see the price before you commit. New numbers are charged straight to your balance: a one-time setup fee per number, plus the monthly rental described below. Once bought, a number lands in My numbers ready to use — already able to answer inbound calls, with its own INBOUND +1… campaign waiting for a prompt — and stays yours until you release it. Buying does not cost you the free shared line: it stays in My numbers alongside the new one, so you always have a caller ID to fall back on.

Dedicated numbers. A number that is yours alone can answer inbound and always dials as itself. Each one carries a small monthly rental drawn from your balance the same way calls are, and it shows in your history as Number rental. The first month is the one you paid for when you bought it, so the next charge falls a month after the purchase. The free shared number is never charged. Should the balance sit too low when a rental falls due, that number is marked SUSPENDED under My numbers and stops placing or receiving calls — your free shared number keeps working — until you top up, at which point it reactivates on its own.

One field does not inherit: a campaign’s system prompt. Calls to real numbers will not start until you have written one, so a half-configured campaign can never dial a customer with a generic assistant. Test calls from the browser are exempt. A campaign’s Outbound Number is also its own: leave it on Automatic and Boltun picks the caller ID closest to each contact, or pin one of your numbers to use it for every call in that campaign.