@kithjs/elizaos

Drop-in voice plugin for ElizaOS agents. Replaces @elizaos/plugin-elevenlabs with Kith's full voice pipeline.

Install

bun add @kithjs/elizaos

Register

In your agent's character file:

import kithVoice from "@kithjs/elizaos";

export default {
  name: "MyAgent",
  plugins: [kithVoice],
  // ...
};

Configure

In your .env:

ELEVENLABS_XI_API_KEY=sk_your_key
ELEVENLABS_VOICE_ID=kPzsL2i3teMYv0FxEYQ6

What It Replaces

Feature	plugin-elevenlabs	@kithjs/elizaos
Sentence chunking	No (raw text dump)	Yes (natural boundaries)
Slang expansion	No	4 built-in dicts
Pronunciation	No	Per-character overrides
Emoji handling	Read aloud	Stripped, emotion events
Laugh tags	No	v3 [laughs] / [giggles]
Multi-provider fallback	No	ElevenLabs + OpenAI + Cartesia
Character profiles	No	VoiceCharacter JSON
Text transforms	No	Custom pipeline

No Python Required

Unlike the full Kith stack (which uses a Python Pipecat sidecar), the ElizaOS plugin uses ElevenLabsDirectAdapter — a lightweight TypeScript-native adapter that calls the ElevenLabs SDK directly. No Python, no sidecar, no extra processes.

Standalone Use

You can use the synthesis function without ElizaOS:

import { synthesizeWithKith } from "@kithjs/elizaos";

const { audio, emotions } = await synthesizeWithKith(
  "Hey! That was fire 🔥 lol ngl",
  {
    apiKey: process.env.ELEVENLABS_XI_API_KEY,
    voiceId: "kPzsL2i3teMYv0FxEYQ6",
    modelId: "eleven_v3",
    enableSlang: true,
    enableEmoji: true,
  },
);
// audio: KithEvent[] with tts_audio_chunk events
// emotions: KithEvent[] with emotion_state events

Environment Variables

Variable	Default	Description
ELEVENLABS_XI_API_KEY	—	ElevenLabs API key (ElizaOS convention)
ELEVENLABS_VOICE_ID	—	Voice ID
ELEVENLABS_MODEL_ID	eleven_v3	TTS model
KITH_ENABLE_SLANG	true	Enable slang expansion
KITH_ENABLE_EMOJI	true	Enable emoji-to-emotion
KITH_CHARACTER_FILE	—	Path to VoiceCharacter JSON