Voice SDK v2 powers natural, low-latency voice agents, sub-300ms responses, native multilingual switching and the tools to embed voice anywhere.
Sub-300ms first response with natural turn-taking and barge-in, no robotic pauses.
Detects and switches language mid-call, so every caller is understood.
Drop voice into calls, your app or the web with a clean SDK and client tools.
Books, qualifies and logs to your CRM through the same agent brain as every channel.
Voice SDK is the browser version of the MyChatBot Calls channel. Instead of a phone number or carrier, you choose the Web SDK provider, and visitors click a button on your site to talk to your AI agent through their own microphone and speakers.
It is the same agent that answers your chats: it follows your instructions and Knowledge Base, speaks in the voice you choose, and each call becomes a conversation with a transcript and a recording.
Typical uses:
It suits businesses that want voice on the web without telephony, and developers who want the agent in their own interface. If you need an agent that answers or places phone calls, see AI Voice Agent, which runs on the same Calls channel with SIP Trunk or Twilio.
Each visitor’s browser gets its own caller identity, so conversations from different visitors never merge. During the call, the button becomes a compact bar showing who is speaking, with mute, end call and a live transcript button.
In the Calls panel you choose a Voice with preview, the LLM behind the answers, Speech Speed from 0.7 to 1.2, the First Message and a Max Call Duration, which is 10 minutes by default. The voice also sets the agent’s primary language. Additional languages adds languages the agent can switch to mid-call, with an optional greeting for each, and Automatic language switching detects the caller’s language.
With client_tools you expose page functions the agent may call while it talks. Each tool has a definition the agent reads, with a description and parameters of type string, number or boolean, and a handler you write. Whatever the handler returns is spoken back, so a short confirmation works best.
The embed brings its own styles. You can set the agent’s name, an accent colour, a bottom-right or bottom-left position, render the control inside a chosen element, and show the transcript collapsed, open or off. The onCallStart and onCallEnd callbacks tell your code when a call connects and ends, with the reason and duration in seconds. For JavaScript builds, the npm package @mychatbot/client lets you start calls from your own button and interface.
Voice SDK is one of three providers of the Calls channel, next to SIP Trunk and Twilio Account. Web calls show up on the Chats screen just like text conversations, so your team reviews them in one place.
Knowledge. The voice agent is built from your agent’s instructions and Knowledge Base when you connect the channel. Load your catalogue into the Knowledge Base, and the agent can answer stock, price and product questions during the call. Because the caller talks to a sales assistant, the conversation can also create a lead to nurture later.
CRM. If you configure Post-Call Functions, they run actions after a call, such as writing a lead to your CRM. The CRMs MyChatBot connects to, including KeyCRM, Bitrix24, Kommo and HubSpot, are listed on the integrations page.
Claude. If you set up your agent through the MyChatBot connector in Claude, ask it for the embed code, and Claude returns the snippet with your real Agent ID.
Web SDK calls run in the visitor’s browser, with these limits:
For a guided start, choose Create a voice agent for my website in the Sales Platform Quick start. Give it your site or product feed URL and a short brief. It sets up a sales assistant with a voice and a Web Voice channel, and returns the Agent ID and an embed snippet with sample page actions: searchItem, navigateTo, addToCartById and applyDiscountCode.
Web calls are billed by duration in AI messages at the same rates as phone calls: high-quality voices use more AI messages per minute than standard voices, prorated by the second with no per-call minimum. Voice SDK plans are on the pricing page.
No. The Web SDK provider needs no phone number, SIP trunk or Twilio account. You pick a voice, click Create SDK Agent and paste the embed with your Agent ID into your site. Visitors talk to the agent in the browser with their microphone.
Web SDK calls are billed by duration and converted into your plan’s AI messages, at the same rates as phone calls. High-quality voices use more AI messages per minute than standard voices, prorated by the second with no per-call minimum. Adding extra languages moves the agent to the high-quality rate.
Yes, through client tools. You describe a page function and write its handler, and the agent can call it mid-conversation, for example to search products, add an item to the cart or apply a discount code. The handler’s return value is spoken back to the visitor.
Yes. The embed can float in the bottom-right or bottom-left corner in your accent colour, or render inside an element you choose. For full control, the npm package @mychatbot/client lets you start and manage calls from your own interface.
The voice you choose sets the primary language. You can add languages the agent switches to mid-call, each with an optional greeting, and turn on automatic language switching to detect the caller’s language. Additional languages use the higher-quality v3 Conversational model, which bills at the higher rate.
Each call is written into the client’s chat with a full transcript and a playable recording, on the Chats screen alongside your text conversations. Every visitor’s browser gets its own caller identity, so conversations do not merge.
No. Transfer to a human operator, auto-recall and outbound dialing are available only for the phone providers, SIP Trunk and Twilio. With the Web SDK, your team follows up from the conversation in Chats.
Grab the SDK, connect your Knowledge Base and tools, and ship a natural voice agent.
Start free