Open source · MIT License

The AI streamer that feels alive.

Design a personality. Point it at your screen. Wallie runs the show — with real opinions, real reactions, and a voice that doesn't repeat itself for hours.

Star on GitHub Watch Demo

Every AI streamer breaks after 10 minutes.

They describe your screen like a help desk. They repeat themselves constantly. They ask "what do you think, chat?" into the void. They're tech demos, not shows.

The usual

Robotic narration

"I can see a YouTube homepage with several video thumbnails arranged in a grid layout."

Wallie

First-person ownership

"okay let me pull this up, I saw someone post about this earlier and I have thoughts."

The usual

Constant repetition

Says the same observation with slightly different words every 30 seconds.

Wallie

Dedupe engine

Bigram/trigram similarity scoring, phrase cooldowns, and rolling history injection catch every repeat.

The usual

Two sentences, then silence

Awkward 3-second gaps between every thought. Choppy, robotic pacing.

Wallie

Organic pacing

Mood-driven tempo, pipeline overlap, silence beats, and thought development across minutes.

Everything a real streamer needs.

One framework. Infinite characters. Every component is swappable.

🎭

Full persona design

Name, energy, humor style, catchphrases, running gags, banned words, opinions, backstory. Not just a system prompt.

👁

Vision that reacts

Screen capture with activity classification. Knows when you're gaming, browsing, or watching a video. Reacts like it's playing.

🧠

Hours without decay

Rolling summarizer compresses old context. Cross-session memory persists facts. No degradation over time.

🎤

Viseme lip sync

Spectral FFT analysis shapes the avatar's mouth per sound. Not just open/close — actual vowel shapes in real-time.

💬

Multi-platform chat

Twitch, YouTube, and Kick simultaneously. Highlights (subs, donations, raids) always get acknowledged.

Browser dashboard

Configure everything from a web UI. No YAML files, no terminal after setup. Live preview and instant testing.

5 LLM providers. 3 TTS engines.
Mix and match per profile.

Language Models

OpenAI Anthropic Gemini Groq OpenRouter

Text-to-Speech

Fish Audio ElevenLabs Piper (free)

Chat Platforms

Twitch YouTube Kick

Avatar

VTube Studio (Live2D)

Design your streamer in the browser.

Identity, personality, voice, topics, vision, chat, avatar, engine — every parameter lives in one dashboard. Save profiles, switch characters, go live.

Wallie Dashboard — Identity configuration

Identity — who your streamer is

Wallie Dashboard — Personality configuration

Personality — how they talk, not what they talk about

From $0 to premium.
Your keys, your cost.

Wallie runs on your machine with your API keys. No subscription, no middleman.

Free

$0 / hour
Great for testing
LLM: Gemini 2.5 Flash
TTS: Piper (local)
Vision: Gemini
Chat: all platforms

Premium

~$6.50 / hour
Best quality
LLM: Claude Sonnet
TTS: ElevenLabs
Vision: Claude
Chat: all platforms

Three commands. One browser tab.

Clone, install, launch. Everything else happens in the dashboard.

1

Clone

Grab the repo from GitHub.

2

Install

Double-click install.bat (Windows) or run start.sh (Mac/Linux).

3

Design & go live

Open the dashboard, design your character, add API keys, hit Start.

# clone & run
git clone https://github.com/Alradyin/wallie-V2.git
cd wallie-V2
# Windows: double-click start.bat
# Mac/Linux:
./start.sh

Built for people who want a real show,
not a tech demo.

Star the repo. Try the free path. Tell us what broke.