Ear3 for developers

Build voice interviews into your product. Embed a component, receive a signed webhook, ship. Or stress-test your interview with AI personas before it ever reaches a real user.

The Ear3 ecosystem

@ear3/voice-interviewer + @ear3/server. Drop <Ear3VoiceInterview> into your web app, verify webhooks server-side. From zero to a live voice interview in 5 minutes — branded widget or headless RTVI with full UI control.

SDK

How the voice pipeline runs. Pipecat Cloud (managed) or self-host with our open Docker image. Same code path, different substrate.

Server

REST endpoints, webhook event catalog, error codes. What the SDK calls under the hood — reach for it when a client library isn’t an option.

API Reference

How the voice pipeline works

Every live interview flows through the same six stages. Ear3 orchestrates them via Pipecat so latency stays under a conversational threshold end-to-end.

   Respondent (browser)
        │
        │  WebRTC audio, ~200 ms glass-to-glass
        ▼
   Daily transport ──► Deepgram STT ──► LLM (OpenAI / Gemini / Anthropic)
                                                    │
                                                    ▼
                              Cartesia TTS ──► WebRTC ──► Respondent
  • Transport — Daily WebRTC rooms for sub-300 ms audio in and out.
  • STT — Deepgram Nova-3 for streaming transcription with word timings.
  • LLM — swappable per-interview. OpenAI GPT-4-class, Anthropic Claude, or Google Gemini — configured per workspace.
  • TTS — Cartesia Sonic for low-latency neural voices; ElevenLabs available for expressive fallback.
  • VAD — Silero for endpoint detection so the model knows when the respondent is done speaking.
  • Orchestration — Pipecat 1.3 connects the pieces and streams frames between them.

Two ways to run it

We run the voice pipeline on Pipecat Cloud. You get warm workers, autoscaling, and multi-region routing out of the box. This is what app.ear3.ai uses in production — nothing to install.

Pipecat Cloud (default)

Prefer to keep audio in your own VPC? Deploy the same pipecat-cloud-server Docker image on any Kubernetes cluster and point Ear3 at your endpoint. Full control, same SDKs.

Self-host

Pick your path

I want to…Start here
Ship a voice interview in my app tomorrowSDK → Quickstart
Understand the mental model (keys, sessions)SDK → Concepts
Run the voice pipeline on my own infraServer → Self-host
Call the API without an npm dependencyAPI Reference

Community

Questions, feedback, feature requests welcome. We read everything.

  • GitHub — github.com/ear3-ai — issues, PRs, discussions
  • Discord — real-time support + roadmap conversation
  • Blog — ear3.ai/blog — launches + engineering notes
Built by Ear3 — voice interviews for any app.
⌘/