What is zoxaAI?
zoxaAI is a managed voice AI platform for building and deploying conversational agents that make and receive phone calls, run browser-based voice sessions, and execute outbound campaigns at scale.
Build voice AI agents in minutes
zoxaAI handles the entire real-time audio pipeline for you -- speech recognition, language model reasoning, and speech synthesis -- so you can focus on what your agent says, not how it talks. No API keys to configure, no infrastructure to manage.
zoxaAI is a managed voice AI platform. You describe how your agent should behave, pick a voice and language model, and zoxaAI runs the entire real-time conversation pipeline at sub-second latency. Your agents can answer phone calls, make outbound calls, run in the browser, or dial through thousands of contacts as part of an automated campaign.
You do not need to bring your own API keys. zoxaAI provides managed access to 4 LLM providers (OpenAI, Google, Anthropic, and Qwen), 6 TTS providers (ElevenLabs, Cartesia, Sarvam, and more), and 5 STT providers -- all included with your account. Pick the best provider for your use case, and zoxaAI handles authentication, billing, and failover.
How the pipeline works
Every conversation follows the same low-latency loop. The entire cycle typically completes in under one second from the moment the caller finishes speaking.
Caller speaks
The caller says something into their phone or browser microphone. Audio streams to zoxaAI in real time.
Speech-to-Text (STT)
zoxaAI transcribes the audio into text using your configured STT provider. Supported providers are Soniox, Deepgram Flux, AssemblyAI, Cartesia, and Sarvam.
LLM reasoning
The transcript is sent to your configured language model along with the system prompt, conversation history, and any retrieved knowledge base content. The LLM generates the next response. If a tool call is needed (HTTP API, call transfer, end call), the LLM triggers it before responding.
Text-to-Speech (TTS)
The LLM's response is synthesized into natural-sounding audio using your configured TTS provider. Supported providers include ElevenLabs, Cartesia, Smallest, Sarvam, Inworld, and xAI.
Caller hears the reply
The synthesized audio streams back to the caller. The loop repeats for every turn of the conversation until the call ends.
When the call ends, zoxaAI saves the full transcript, audio recording, cost breakdown (LLM tokens, TTS characters, STT minutes, telephony minutes), and metadata -- all accessible from the dashboard or the API.
What you can build
zoxaAI is designed for production voice AI across industries. Here are specific use cases the platform supports today:
| Use case | How it works |
|---|---|
| Customer support line | Attach an agent to an inbound phone number. It answers calls, troubleshoots issues, looks up account data via HTTP tools, and escalates to a human with call transfer when needed. |
| Outbound sales cadence | Create a campaign with a CSV of leads. The agent calls each contact, qualifies interest, answers objections, and books meetings by checking calendar availability via an HTTP tool. |
| Appointment scheduling | An agent on your phone line collects date/time preferences, checks availability through your booking API, and confirms the appointment. |
| Surveys and feedback | Upload a contact list to a campaign. The agent calls each person, asks a structured set of questions, and the call record is sent to your system via the agent's webhook. |
| Lead qualification | The agent qualifies leads by asking budget, timeline, and use case questions, then transfers hot leads to your sales team. |
| Payment reminders | Campaign dials overdue accounts. The agent confirms identity, states the balance, offers payment options, and can transfer to a payment line. |
| AI receptionist | An inbound agent answers your business phone number 24/7, handles FAQs using a knowledge base, collects caller information, and transfers to the right department. |
How agents work
An agent is the fastest way to get a voice AI live. You write a system prompt that describes how the agent should behave, pick a voice and language model, and optionally attach tools and a knowledge base. The agent holds a free-form conversation within your instructions, which suits open-ended topics like customer support or general inquiries as well as structured scripts you describe step by step in the prompt.
You configure an agent in a form-based editor with five tabs (Agent, Engine, Tools, Call, and Analytics) plus a right panel where you pick the transcriber, model, and voice. Tools are attached to the agent and available throughout the call. When the call ends, the call record (transcript, recording, cost) is saved and can be sent to your backend through the agent's webhook.
Three ways to connect
zoxaAI supports multiple transports so your agents can reach users wherever they are.
| Transport | How it works | Phone number required |
|---|---|---|
| Web Call (WebRTC) | Voice call runs directly in the browser. No phone needed -- the user clicks a button and talks through their microphone. Ideal for testing and in-app voice experiences. | No |
| Phone Call (Telephony) | Standard telephone call via a connected telephony provider. Supports both inbound (caller dials your number) and outbound (you trigger a call via the dashboard or API). | Yes |
| Campaign Call | Automated outbound telephony call to a contact from a batch campaign. zoxaAI dials through a CSV list with concurrency control, retries, and scheduling. | Yes |
All three transports produce the same call record when they end -- transcript, recording, and cost breakdown. Your analytics and reporting are consistent regardless of how the call was initiated.
Start with web calls
Web calls are the fastest way to test your agent. No phone number, no telephony provider, no setup -- just click "Web Call" on any agent page and start talking.
Platform highlights
| Feature | Details |
|---|---|
| Managed API keys | All LLM, TTS, and STT providers are included. No need to create accounts with OpenAI, ElevenLabs, or Soniox separately. |
| Sub-second latency | The STT-LLM-TTS pipeline typically responds in under one second. |
| 4 LLM providers | OpenAI, Google Gemini, Anthropic Claude, and Qwen (Alibaba) -- all keys managed by the platform. |
| 6 TTS providers | ElevenLabs, Cartesia, Smallest, Sarvam, Inworld, and xAI. |
| 5 STT providers | Soniox, Deepgram Flux, AssemblyAI, Cartesia, and Sarvam. |
| 8 telephony providers | Twilio, Vonage, Telnyx, Plivo, Cloudonix, VoBiz, Tata Smartflo, and Asterisk ARI. |
| Knowledge base (RAG) | Upload PDF, DOCX, TXT, or JSON files. The agent retrieves relevant content during calls. |
| Campaign engine | Upload a CSV and dial thousands of contacts with concurrency control, retry logic, scheduling, and circuit breaker protection. |
| Full call records | Every call saves a transcript, recording, cost breakdown, and metadata. |
| API and webhooks | Trigger calls, manage agents, and receive call events programmatically. |
| Analytics and tracing | Monitor agent performance with call analytics, latency tracing, and cost tracking across all your agents. |
| Multilingual support | Many STT and TTS providers support 30+ languages. Build agents that speak to callers in their preferred language. |
Get started
Quickstart
Create your first agent and make a test call in under 5 minutes
Key Concepts
Understand agents, calls, runs, campaigns, tools, and knowledge bases
API Reference
Trigger calls and manage agents programmatically
Agents
Full configuration reference for agents
Campaigns
Run outbound calling campaigns at scale