Roadmap

Our Mission

Build AI companions you actually own.

We believe everyone should be able to create an AI companion that runs on their hardware, speaks with a voice they choose, remembers their conversations, and answers to them — not a corporation. No API bills. No data leaving your machine. No company deciding what your companion can or can't say.

Kith is the open-source toolkit that makes this possible.

What's Shipped

v0.1 — Foundation

The core voice quality layer. 10 npm packages under @kithjs/*.

Core contracts — 5 adapter interfaces, 12-event normalized bus
Pipecat runtime — Python sidecar with ElevenLabs TTS, auto-respawn, barge-in
Voice router — Sentence-aware chunking, 4 slang dicts, pronunciation, emoji-to-emotion, VoiceCharacter profiles, laugh tags
LiveKit adapter — WebRTC production adapter + mock mode
Observability — Traces, dup-send guards, reconnect metrics
Avatar events — Viseme-to-blend-shape mapping, expression state machine

v0.2 — Multi-Provider + Framework Plugins

Production reliability and framework integrations.

Multi-provider TTS fallback — ElevenLabs + OpenAI TTS + Cartesia with circuit breaker
Mic input / STT — Whisper + Deepgram, VAD silence detection
OpenClaw plugin — openclaw plugins install @kithjs/openclaw
React Native support — LiveKit RN SDK helpers

v0.3 — Standalone Server + ElizaOS

Drop-in voice for any framework.

@kithjs/server — Standalone voice microservice (REST + WS + SSE + Docker)
@kithjs/client — TypeScript client for the server
@kithjs/elizaos — Drop-in replacement for plugin-elevenlabs with ElevenLabsDirectAdapter (no Python needed)
Local TTS pipelines — Piper (CPU, runs on Raspberry Pi) and Voxtral (GPU, voice cloning from 5 seconds of audio)

What's Next

v0.4 — Own Your Companion

The local-first companion stack. No cloud APIs required.

Feature	Description	Status
Local TTS (Piper)	CPU-only TTS that runs on any machine. No GPU, no API key, no internet.	Pipeline built
Local TTS (Voxtral)	GPU-accelerated TTS with voice cloning from 3-5 seconds of audio. Your companion, your voice.	Pipeline built
Ollama adapter	Connect VoiceRouter to local LLMs via Ollama. Full voice conversation with zero cloud dependency.	Next
@kithjs/companion	Generalized relationship system — progression (stranger → bestie), memory, personality adaptation, greeting engine. From the Kaori integration, available to all.	Next
Emotion-aware voice	Detected emotions feed back into TTS parameters. Excited text → higher energy voice. Sad text → softer tone.	Next
Character creator UI	Web interface: name your companion, describe their personality, record voice for cloning, export VoiceCharacter JSON. No code needed.	Planned

v0.5 — One-Click Companion

Feature	Description
`docker run kithjs/companion`	Single Docker image: Ollama + Piper/Voxtral + Kith server + web UI. Your companion runs on your machine, fully offline.
Open-source STT	Local Whisper (whisper.cpp) replacing the cloud Whisper API for fully offline STT.
Conversation export	Export your companion's memory and conversations. Your data, always portable.
Multi-companion	Run multiple characters with different personalities on the same server.

Future

Feature	Description
Telephony	SIP/PSTN via LiveKit — call your companion from a phone
Mobile app	React Native companion app with local or cloud voice
Federated memory	Share companion knowledge between instances without centralizing data
Plugin marketplace	Community-built character profiles, voices, and transforms

The Stack (Where We're Heading)

┌─────────────────────────────────────────┐
│              Your Companion              │
│  Character · Personality · Memory · Voice│
└──────────┬──────────────────┬───────────┘
           │                  │
    ┌──────▼──────┐   ┌──────▼──────┐
    │  Brain      │   │  Voice      │
    │  (your LLM) │   │  (Kith)     │
    │             │   │             │
    │ Ollama      │   │ VoiceRouter │
    │ Claude      │   │ + TTS       │
    │ GPT         │   │ + Emotion   │
    │ Llama       │   │ + Avatar    │
    └─────────────┘   └─────────────┘
           │                  │
    ┌──────▼──────┐   ┌──────▼──────┐
    │  Cloud      │   │  Local      │
    │  (optional) │   │  (default)  │
    │             │   │             │
    │ OpenRouter  │   │ Piper (CPU) │
    │ Anthropic   │   │ Voxtral     │
    │ OpenAI      │   │ Whisper.cpp │
    └─────────────┘   └─────────────┘

Cloud is optional. Local is the default. Use cloud providers when you want the best quality. Use local when you want privacy, zero cost, and full control.

Philosophy

Your companion, your data. Conversations never leave your machine unless you choose to use a cloud provider.
No vendor lock-in. Swap TTS providers, LLMs, or runtimes without changing your companion code.
Accessible to everyone. You shouldn't need to be a developer to create a companion. Character creator, one-click deploy, no-code setup.
Open source forever. MIT licensed. The core toolkit will always be free.
Emotion is not optional. A companion that sounds robotic isn't a companion. Natural voice, real laughter, emotional awareness — these are table stakes.

Contributing

We welcome contributions at every level:

Local TTS providers — Kokoro, Fish Speech, MOSS-TTS, Qwen3-TTS
Ollama / local LLM integration
Character profiles — share your companion personalities
Framework plugins — LangGraph, CrewAI, AutoGen adapters
Voice models — cloned voices, accent support
Documentation — guides, tutorials, translations
Bug reports — from real deployments

GitHub: github.com/wbaxterh/kith

Our Mission​

What's Shipped​

v0.1 — Foundation​

v0.2 — Multi-Provider + Framework Plugins​

v0.3 — Standalone Server + ElizaOS​

What's Next​

v0.4 — Own Your Companion​

v0.5 — One-Click Companion​

Future​

The Stack (Where We're Heading)​

Philosophy​

Contributing​