open-yoki
An AI live-stream host that runs on one server: voice, face, chat, memory, posts. The base yoki runs on. By Jacke_pp · contact: DM @Jacke_pp on X
- Stars
- 213
- Forks
- 1
- + today
- +6
- Created
- 9d
Ranking data as of November 5, 2026 (UTC).
Star History
Today, hour by hour
Overview
An AI live-stream host that runs on one server: voice, face, chat, memory, posts. The base yoki runs on. By Jacke_pp · contact: DM @Jacke_pp on X It ranks #79 on GitTiger, gaining +6 stars on November 5, 2026 (UTC).
The project is written in JavaScript and has 1 fork. It was created 9d ago.
What it is
open-yoki is an AI live-stream host designed to run on a single server. The description sums it up as voice, face, chat, memory and posts in one package. The repository is JavaScript and MIT-licensed.
Capabilities described
The streamer appears with a cloned voice and a real-time digital-human visual, reads incoming chat, replies, remembers returning viewers, keeps a diary, publishes posts on its own, and starts or stops broadcasting on a schedule — both to Binance Square (or anywhere accepting RTMP) and to its own web live room.
Mechanically, the Chinese-language preview describes a headless Chrome instance drawing a 16:9 frame with the character, subtitles, a market panel, thanks for gifts and follows, music and sound effects, pushed out over RTMP with ffmpeg during the broadcast window. Replies are handled by a fast model — DeepSeek through OpenRouter by default — answering within seconds and prefixing filler sounds such as "mm-hm" while it thinks, with mood, actions and tone following what it says.
Getting started
No installation steps were supplied; the preview references internal scripts and a skill document rather than a setup guide. Read the README for prerequisites and how to point it at a stream target.
Caveats
Everything above, including latency and quality claims, comes from the project's own README.
git clone https://github.com/jacke-pp/open-yoki.git
cd open-yoki
# see README for setupFrequently Asked Questions
What can the AI host do on its own?
Per the preview: read and answer chat, remember regulars, write diary entries, publish posts, and start or stop streaming on a timetable.
Where can it stream?
To Binance Square or any endpoint that accepts RTMP, and to its own web live room.
Which model generates the replies?
A fast model, DeepSeek via OpenRouter by default, answering chat within seconds according to the preview.
What does the on-screen frame include?
Per the preview: the character, subtitles, a market panel, thanks for gifts and follows, music and sound effects, rendered at 16:9 by headless Chrome.