An AI customer service agent that answers "where is my order?" from the connected store, "how do I return it?" from your own help docs, and "do you ship to Ireland?" from the data points you gave it, on email, website chat, texts, social DMs and the phone. The same answer on every channel, every reply kept with its source, and the refund handed to a person.
14-day free trial · No credit card · Cancel anytime
Connect the support@ mailbox your customers already write to. Gmail and Outlook need an app password; any other provider goes in as an SMTP and IMAP host. New mail comes in over IMAP, the agent reads it, and the answer goes back from the same address through your own server, so the thread stays on your domain and inside the customer's existing conversation.
It replies once per thread and ignores bounces, out-of-office notes and no-reply addresses, so an auto-responder never starts a loop. When the customer writes back with a follow-up question, that is a genuine message and it is answered. When the customer writes back asking for a refund, that is the moment it stops and fetches you.
One line of script puts a widget on your site in your colour, with your image and a greeting you wrote. A visitor asks whether the blue one comes in a large, how long delivery takes, or what the returns window is, and the agent answers from your documents, your Q&A pairs and your data points, linking only to the pages of your own site you allowed.
When the conversation is worth keeping it asks for an email address, and that address ties the visitor to the rest of their history. The next time they email or text, the agent already knows what they asked on the site, and the person who takes the thread over can read it from the top.
Messenger, Instagram, WhatsApp, Telegram, LINE, Viber, Slack, Discord, Bluesky, Reddit and X, each connected with your own app credentials and answered by the same agent inside each platform's reply window. Texts run on your own number through Twilio, Telnyx or one of seven other carriers, and support calls are answered on Twilio with the recording and transcript kept.
Because it is one agent with one knowledge base, the returns policy is the same in a WhatsApp message as it is in an email, and the customer who DMs at 11pm gets the same answer as the one who called at 11am. Auto-reply is a separate switch per channel, so you can switch it on for chat and DMs first and add email when you are ready.
Email, chat, texts, DMs and call transcripts arrive in one inbox, newest first, each row tagged with the channel it came in on. Filter it to a channel, a number, a mailbox or an agent. Every reply the agent wrote is stored beside the document, Q&A pair or data point it was generated from, so you can read back exactly why it said what it said.
Reply in any thread yourself and the agent stands down on that conversation until you hand it back. That is how refunds, complaints and anything else you named in its instructions get to a person: not through a ticket queue, but in the same thread, with the whole history above the reply box.
Most support questions are the same twenty questions. Where is my order. How do I return it. Do you ship there. What are your hours. Is it in stock. Every one of them has an answer already written down somewhere in your business: the help centre, the FAQ page, the returns policy, the order record. The problem is not that the answers do not exist. It is that a person has to find them and type them out, one at a time, for every customer, including the ones who write at 11pm on a Sunday.
Bold Agent Kit gives you one AI customer service agent that answers those questions on every channel a customer might use: the support@ mailbox, the chat widget on your website, texts to your number, DMs on Instagram, Messenger, WhatsApp and eight other platforms, and the phone. It answers from the material you gave it, your documents, your Q&A pairs, your prices, hours and lead times, and the pages of your own site it may link to, and from nothing else. What it says is what you wrote, and every reply is stored alongside the source it came from. That is what automated customer service should mean: your answers, delivered by software, not a chatbot improvising.
Connect your store and it reads the order before it writes a word. With Shopify, WooCommerce, Wix, Squarespace or Stripe connected, "where is my order?" is answered with what is actually on that customer's record, not with a generic "please allow five to seven working days". It sees what they bought, what is still open and what they are worth, and it treats the email, the DM and the chat from the same person as one conversation, because it merges them into one contact before it replies.
And it knows what it is not for. A refund, an exchange, a complaint, a chargeback: you write those into its instructions as the moments to stop and fetch you, and when a message matches, the thread lands in your inbox with the agent standing down until you hand it back. It cannot issue a refund or change an order itself, and that is deliberate. An AI support agent takes the twenty questions off your desk so the five that need a person get one, and get one sooner.
Customers are not hard to please. They want a straight answer to a plain question, soon. Four things stand between them and it.
The returns window is on the policy page. The delivery time is in the shipping FAQ. The order status is in the store admin. A person opens three tabs, copies the answer, pastes it into a reply, and does it again for the next customer, who asked the same thing ten minutes later.
Support questions arrive when people are at home with their order, not between nine and five. An email sent on Saturday morning is read on Monday, by which time the customer has asked twice more, on two more channels, and has started to wonder whether anyone is there at all.
The person on chat says fourteen days. The person on email says thirty. The Instagram DMs are answered by whoever has the phone, from memory. The customer who asks in two places gets two answers and believes neither of them.
When every message is treated the same, the complaint that needs a person waits in the same queue as the question that needed a link. The easy ones make the hard ones late, and the hard ones are the ones that cost you a customer.
You build the customer service agent once, from the material you already have. From then on it answers on every channel you connect, on its own, and only fetches you for the conversations you told it to.
Upload the help centre, the returns policy and the shipping terms as PDF or Word. Add Q&A pairs for the questions you get every day, your prices, hours and lead times as data points, and the pages of your own site it may link to. Every reply on every channel is generated from that material and stored beside it, so you can always see why it said what it said.
With Shopify, WooCommerce, Wix, Squarespace, Stripe or your CRM connected, it looks the customer up before it writes. What they bought, what is still open, what they are worth. "Where is my order?" is answered from the record, and the email, the DM and the chat from one person are read as one history, because they are one contact.
Refunds, exchanges, complaints, anything you name in its instructions. When a message matches, the thread lands in your inbox with the full history and the agent standing down on that conversation. On a phone call, it can transfer the live call to a person on your team instead. You reply in the same thread, from the same address, and hand it back when you are done. It never talks over you.
Every thread has a reply box for you. Type in it and the agent stands down on that conversation, so it never talks over you or answers a question you are halfway through handling. The customer sees one continuous conversation, first with the agent and then with you, on the same channel and from the same address or number. When you are finished, hand the thread back and the agent picks it up where you left it.
The handover is also how the difficult ones reach you without a ticket queue. A refund request, a complaint, a customer who has written three times: name them in the instructions and they come to your inbox with the whole history above the reply box and the source of every answer the agent gave along the way. That is 24/7 customer support automation that knows its own limits.
Each one answers a problem above, and each one runs without anybody remembering to do it.
A question at 11pm is answered at 11pm, on the channel it arrived on, from your help docs. No queue, no working hours, no "we aim to respond within two business days".
The returns window is the same in a WhatsApp DM as it is in an email, because the same agent with the same knowledge base wrote both. Change the policy once and every channel follows.
It reads the connected store and the customer's whole history before it replies, so nobody is asked for their order number twice and nobody has to repeat themselves.
The twenty routine questions are handled. The refund, the complaint and anything else you named come to a person, in the same thread, with the history attached.
Every message, reply, call and visit is recorded against a person and reported per agent and across all agents. This is the part that tells you which channel carries your support load, and where a support conversation turned into a sale.
The funnel counts people, not tickets: how many the agent reached, how many replied, how many went on to visit your site, and how many converted, with the value won and the median time it took. A support agent is often the last thing a customer talks to before they buy again, and a goal with a dollar value attached tells you how often that happens and what it was worth.
Below the funnel, a channel table shows where each person first made contact and which channel they were on when they converted, so you learn whether it is the chat widget or the Instagram DMs that your customers actually use. One row per person shows their whole history in order, across every channel, and visit triggers close the loop: when a known customer opens the returns page and leaves, the agent can tell you, or follow up after a delay on the channel they last used.
Built from the cookie, email address and phone number attached to that contact, so activity months apart on different channels appears on one row.
Page goals and visit triggers require a one-line snippet on your website. Conversations, bookings and manually marking a deal won work without it.
Each of these is usually a separate tool or a separate person, and most of them are priced per seat, per ticket, per conversation or per resolution. Here they are one plan.
| The job | What it normally takes | Bold Agent Kit |
|---|---|---|
| Answering email at nights and weekends | A second shift, an outsourced help desk, or Monday | Included |
| Live chat on your website | A chat vendor priced per conversation or per resolution | Included |
| Instagram, Messenger and WhatsApp DMs | Whoever has the phone, or another inbox tool | Included |
| Support calls | A phone answering service, priced per minute | Included, on your own number |
| Answering "where is my order?" | A person in the store admin, copying and pasting | Included, read from the connected store |
| The same answer on every channel | A macro library and a training document nobody reads | Included, one knowledge base |
| Getting refunds and complaints to a person | Everything goes to a person | Only the ones you named |
| The AI help desk itself | Metered per seat, per resolution or per credit | Your own key, billed by your provider at cost |
Bold Agent Kit is $25 a month, or $200 a year, for the whole account: every agent, every channel, every conversation, no seat tiers and no per-ticket fee. The only other costs are your AI provider's usage, typically a few dollars a month per agent, and your carrier's for texts and calls, which you would be paying anyway.
The same agent covers all of these. Many accounts run one per brand, one per store, or one per client.
A store on Shopify, WooCommerce, Wix or Squarespace. The agent answers "where is my order?", "can I change the size?" and "what is your returns policy?" from the store record and the help docs, on the chat widget, the support@ mailbox and the Instagram DMs, and hands the refund to a person with the order details already in the thread.
A clinic, a salon, a trade or a firm. Customers ask about hours, prices, availability, parking and what to bring, on the phone, on the website and by text. The agent answers from your data points, books a real slot in the calendar while they are still there, and puts the call transcript in your inbox.
An agency running support for several clients, or a software company with a help centre. One agent per client or product, each with its own mailbox, chat widget, knowledge base and instructions, and nothing shared between them. Invite a client into their own agent and they see their conversations, and nothing else.
Connect the tools you already run and the agent works them like an assistant would: it takes appointments straight into your calendar, writes notes and outcomes back to the CRM, and reads your store to see what somebody has bought before it answers them.
Bookings are real, not guesses. It is only ever given slots that are genuinely free, and it can only take one it actually offered — so it cannot invent a time or double-book you. The same connection pulls a segment of past customers into a list when you want to run a campaign at them.
HubSpot
GoHighLevel
Housecall Pro
ServiceTitan
Jobber
Google Calendar
Microsoft 365
Calendly
HubSpot Meetings
Zoho Bookings
Acuity
Cal.com
Square
Stripe
Shopify
WooCommerce
Wix
SquarespaceEach agent uses a model on your own API key, in the same way it sends from your own mailbox and your own phone number. Sixteen providers are supported: OpenAI, Anthropic, Gemini, xAI and Mistral; DeepSeek, GLM, Kimi, Qwen and MiniMax; Groq, Together, Fireworks and OpenRouter; and Ollama or any other OpenAI-compatible endpoint you host yourself.
Bring your own credentials and you own the whole stack: your model provider, your mailboxes, your carrier, your CRM. We charge for the software and nothing else — no token markup, no metering, no credit packs — so your provider bills you direct at their rates, usually a few dollars a month per agent.
OpenAI
Anthropic
Gemini
xAI
Mistral
DeepSeek
GLM
Kimi
Qwen
MiniMax
Groq
Together
Fireworks
OpenRouter
Ollama
Any OpenAI APIA new address sending forty cold emails a day gets filtered, and takes the domain with it. Mailboxes that opt into warm-up send each other ordinary mail on a ramp, starting at five a day. The receiving mailbox finds each message, moves it out of spam if it landed there, marks it read and replies to a share of them.
Each mailbox carries a health score out of 100 — SPF, DKIM and DMARC on the sending domain, seven days of real sends and bounces, and where its own warm-up mail is landing — with the reasons listed. A mailbox whose bounce rate or spam placement passes a threshold is pulled out of campaign rotation automatically, before it damages the domain.
Three steps. Inbound support never needs a campaign: connect the channels, switch auto-reply on, and let it answer what already arrives.
Name it, choose a model and paste your API key. Then give it what it needs to answer a customer: the help centre, returns policy and shipping terms as documents, Q&A pairs for the questions the documents do not quite cover, your prices, hours and lead times as data points, and the pages of your own site it may link to, which are read, stored and linked verbatim.
Write the instructions last: how it should talk, what it may offer, and the exact situations in which it should stop and fetch you. Refunds, exchanges, complaints and anything involving a payment are the usual four.
Add the support@ mailbox, the chat snippet for your website, a number for texts and calls, and the social accounts your customers actually message. Each one runs on your own account, so the address, the number and the reputation stay yours.
Then connect the store or CRM, so it reads the order before it answers, and a calendar if support ever involves booking somebody in.
Auto-reply is a separate switch for each channel. Switch it on for the chat widget first, read what the agent says in the inbox for a day, then add the DMs, the mailbox and the phone. Calls have their own switches for answering and dialling, so it can answer support calls without ever placing one.
Then watch the inbox. Routine questions are handled; the ones you named come to you, with the history attached. Step into any thread and the agent stands down; hand it back when you are done.
One account runs unlimited agents, one per client, and nothing is shared between them. A knowledge base belongs to a single agent, and content from one agent is never used to generate replies for another.
Invite a client by email address and they get their own login, which opens onto the one agent you gave them: its conversations, its numbers, its analytics and a reply box. They cannot see your other clients, your API keys, your settings or your billing. Access is deny-by-default.
No seat tiers, no credit packs, no per-agent or per-channel pricing, and nothing held back behind an enterprise call.
| Agents | |
|---|---|
| How many | Unlimited — one per business, brand or job |
| Shared brain | The same identity, instructions and knowledge on every channel |
| Knowledge base | Documents, Q&A pairs, data points and pages of your own site |
| Isolation | One agent's content is never used to reply for another |
| Client seats | Invite a client into one agent, with their own login |
| Channels | |
| As many mailboxes per agent as you connect, rotated by campaigns | |
| Text messages | Nine providers — Twilio, Telnyx, SimpleTexting, Plivo, Vonage, Sinch, Bandwidth, MessageBird, ClickSend |
| Phone calls | Answering and dialling, with recordings and transcripts |
| Transcription | Google, or Deepgram nova-3 on your own key |
| Voices | Google, Amazon Polly and ElevenLabs, or your own ElevenLabs / Fish Audio account |
| Voice cloning | Record a voice in the browser and clone it on your ElevenLabs or Fish Audio key |
| Website chat | A widget on your site with your colour, image and greeting |
| Social DMs | Messenger, Instagram, WhatsApp, Telegram, LINE, Viber, Slack, Discord, Bluesky, Reddit, X |
| Inbox | |
| One list | Every channel's conversations together, newest first, tagged by channel |
| Filters | By channel, by mailbox or number, or by agent |
| Takeover | Reply yourself and the agent stands down on that thread |
| Hand back | Return the thread to the agent when you are done |
| Campaigns | |
| Multichannel sequences | Email, text and call steps in one campaign, each with its own timing |
| Send conditions | Send a step only if they opened, did not open, clicked or did not click |
| Schedule | The days, hours and time zone anything may be sent |
| Stop rules | A reply, a booking or an opt-out on any channel ends the sequence |
| Pacing | Per-mailbox and per-number hourly limits, pausable and resumable |
| Review before start | The full plan is shown; a campaign sends nothing until you start it |
| Templates | Email, text and call scripts, saved once and reused |
| Analytics | |
| Funnel | Reached, replied, visited, converted, plus value won and time to convert |
| By channel | Where each person started, and the channel they converted on |
| Pages | Which pages known people read, and how many of them converted |
| Journeys | One row per person: every message, visit, call and booking in order |
| Goals | A page rule or a manual "won", each with a dollar value |
| Triggers | A known person returns to a page — notify you, or send a follow-up |
| Open and click tracking | Recorded per campaign and added to that person's history |
| Integrations | |
| Calendars | Google, Microsoft 365, Calendly, Cal.com, Acuity, Zoho, Square, HubSpot Meetings |
| CRMs and commerce | HubSpot, GoHighLevel, Housecall Pro, ServiceTitan, Jobber, Stripe, Shopify, WooCommerce, Wix, Squarespace |
| Booking | Real open slots offered, and only an offered slot can be booked |
| CRM import | Import a segment of customers into a contact list |
| Deliverability and compliance | |
| Mailbox warm-up | Ramped sending across the network, with spam rescue and replies |
| Health score | SPF, DKIM, DMARC, bounce rate and inbox placement, with reasons |
| Auto-rest | A failing address is removed from rotation automatically |
| Unsubscribes | One click, honoured automatically, including on queued mail |
| Reply limits | One reply per thread, per-sender daily and per-channel hourly caps |
| Filtering | Bounces, out-of-office and no-reply addresses ignored |
| List building | |
| Contact extraction | Emails, phone numbers and social profiles, on any site. Unlimited. |
| Deep scan | Follows the pages of a domain most likely to carry contact details |
| Domain queue | Load a CSV of sites and work through them one at a time |
| Contacts and lists | Unlimited, by email or phone, with CSV import and export |
| AI | |
| Your own key | Sixteen providers, including any OpenAI-compatible endpoint you host |
| What we charge for AI | Nothing — no markup, no metering, no credit packs |
| On calls | Uses your provider's fastest model automatically |
| Account | |
| Platforms | Chrome side panel and web dashboard on one login |
| Members' forum | Six boards, visible only once you are signed in |
| Referral program | $50 a year per referral ($5 a month), recurring for the life of the account |
| Support | support@boldagents.co |
No credit card to start. Billed monthly. No add-ons, no seat tiers. Cancel anytime.
No credit card to start. $16.67 a month, billed once a year. No add-ons, no seat tiers. Cancel anytime.
Anything you have given it an answer to. Upload your help centre, returns policy and shipping terms as documents, add Q&A pairs for the questions you get every day, and enter your prices, hours and lead times as data points. It answers "how do I return this?", "do you ship to Ireland?", "what are your hours?" and "is it in stock?" from that material, on email, website chat, texts, social DMs and the phone, and links only to the pages of your own site you allowed. With your store connected it also answers "where is my order?" from the customer's actual record. Anything it has not been given, it does not answer from memory; it says so and, if you told it to, hands the thread to you.
No, and that is deliberate. It can read a connected store to see what a customer bought, what is still open and what they are worth, and it can answer questions from that record. It cannot issue a refund, cancel an order, change an address or take a payment. Write those as the moments it should stop and fetch you, and when a message matches, the thread lands in your inbox with the whole history above the reply box and the agent standing down on that conversation. You do the refund in your store, reply in the thread, and hand it back when you are done.
From the store you connect. Shopify, WooCommerce, Wix, Squarespace and Stripe are among the eighteen tools it can read, and with one connected it looks the customer up before it replies, so "where is my order?" is answered from what is actually on their record rather than from a generic delivery estimate. It finds them because every handle it has seen for a person, the email address, the phone number, the social account and the browser cookie from the chat widget, is attached to one contact, and when a DM turns out to belong to an existing customer the two records are merged and the history moves across.
It only says what you told it to say. Every reply on every channel is generated from the knowledge base you gave that agent, your documents, Q&A pairs, data points and the pages of your own site, and each reply is stored alongside the material it was generated from, so you can open any thread and see exactly which document or Q&A pair produced the answer. If a customer asks something the material does not cover, the agent does not fill the gap from general knowledge; it says it will pass the question on, and the thread is there for you in the inbox. Content from one agent is never used to answer for another.
Yes. Calls run on your own Twilio number, with separate switches for answering and dialling, so a support agent can pick up every inbound call without ever placing one. It hears and speaks in real time, using the fastest model your provider offers so it does not leave silences, in a voice from Google, Amazon Polly or ElevenLabs, or one you cloned on your own ElevenLabs or Fish Audio key. It can transfer a live call to a person on your team, and the recording and transcript land in the same inbox as everything else, tagged as a call, on that customer's record.
A website chatbot answers on the website. This is one agent with one knowledge base on every channel a customer uses: the chat widget, the support@ mailbox, texts to your number, DMs on eleven platforms and the phone. The returns policy is the same in a WhatsApp message as it is in an email because the same agent wrote both. It also reads your store before it replies, merges the email, the DM and the chat from one person into one contact, hands the threads you named to a person in the same inbox, and reports what happened as a funnel of people rather than a count of chats.
Whatever you wrote in its instructions. Most people name complaints, refund demands and any customer who has written more than twice as the moments to stop and fetch a person, and when a message matches, the thread comes to your inbox with the agent standing down on that conversation. You reply in the thread yourself, on the same channel and from the same address, and the customer sees one continuous conversation. The agent replies once per thread and respects a daily cap per sender and hourly caps per channel, so it never piles automated replies onto somebody who is already upset.
Nothing per ticket, per seat or per resolution. The plan is $25 a month or $200 a year for the whole account, with unlimited agents, contacts and conversations, and a 14-day free trial. The AI runs on your own key with one of sixteen providers, billed to you by that provider at cost, typically a few dollars a month for a busy agent on a small model, with no markup or metering from us. Texts and calls are billed by your own carrier at their rates. A month with three times the support volume costs the same $25.
One subscription, unlimited agents. Most accounts run several, one per job.
Upload the help docs, connect support@ and the chat widget, switch auto-reply on. The routine questions are answered while you sleep, and the ones that need you are waiting in one inbox in the morning.
14-day free trial · No credit card · Cancel anytime