Agent Platform

Speko

View as Markdown

Speko is a voice agent platform for building, evaluating, and monitoring voice agents. Speko lets you configure your TTS and STT stages independently in Cascaded mode — pick Smallest AI’s Lightning v3.1 for voice output, Pulse for speech-to-text, or both.

Prerequisites

  • A Speko account
  • Access to your agent’s Models configuration in the Speko dashboard

Setup

1

Create a Speko account

Sign up at speko.ai if you don’t already have an account.

2

Open the Models tab

In Speko, open the Models tab for your agent and choose Cascaded mode with the Advanced model setup.

3

Set the TTS and/or STT stage

In Cascaded Stages:

  • Set the Voice (TTS) stage to Lightning v3.1, Smallest AI Lightning TTS for speech output.
  • Set the STT (Speech-to-Text) stage to Pulse, Smallest AI Pulse STT for transcription.

You can configure one or both stages with Smallest AI.

4

Save and verify

Save and run your agent, then verify it in Speko’s Evaluation suite with reliability checks and live monitoring.

Notes

  • Speko benchmarks Pulse as its fastest streaming STT option, which keeps end-of-turn latency — the wait between a caller finishing and the agent replying — low.
  • Speko lists Lightning v3.1 (smallest:lightning_v3.1) in its TTS model benchmarks for naturalness and synthesis latency, including English, German, and French.
  • See Speko’s Pulse STT benchmark and Speko’s Lightning v3.1 TTS benchmark for latency and quality numbers.
Speko

Build, evaluate, and monitor voice agents

Speko Documentation

Full Speko platform documentation

Lightning v3.1 Overview

Supported voices, languages, and streaming

Pulse STT Overview

Supported languages, features, and limits