Categoría

Voice agents & realtime AI — phone, support, and live conversation stacks

Vapi, Retell, Bland, LiveKit Agents, Cartesia, Hume, Deepgram, Pipecat and more—build AI that can pick up the phone.

Voice agents exploded in 2025: chain **STT + LLM + TTS + telephony/WebRTC** to run front-desk, pre-sales, scheduling, and follow-ups. Four things decide the build: **end-to-end tail latency** (target < 700ms), **barge-in and turn-taking**, **regional numbering and compliance** (TCPA, two-party consent), and **hosted vs self-built**. On TTS, judge realism and voice-clone licensing; on STT, stress-test noise and accents.

Editorial / GSC

Vapi vs Retell vs Bland

Vapi and Retell are dev-first and let you swap STT/LLM/TTS. Bland is more batteries-included. Record a real script and run 50 calls through each before committing.

How do you get low latency with barge-in?

You need VAD plus duplex audio, streaming STT, streaming LLM, and streaming TTS. LiveKit Agents and Pipecat are popular plumbing; Cartesia and ElevenLabs drive the TTS-side latency that makes conversations feel natural.

Compliance and disclosure duties?

Most jurisdictions require consent before recording/transcription. Outbound phone work also involves numbering rules, holiday/time windows, and do-not-call lists—get local counsel.

Herramientas en esta categoría

Los resúmenes y enlaces oficiales están en cada ficha; navega otras entradas de la misma categoría.