Soomiles is building Animalive for independent streamers: an AI character that follows live chat and responds with a voice, a face, and a distinct personality. Creators control the character while spending less time juggling chat, audio, and an avatar.
Talk to the builderviewer_42: Haru, it's raining here. What are you drinking?
Haru: Hot cocoa, obviously! Rainy days are the best for it. Stay dry out there, okay?
Illustrative interaction, not a live demo.
An LLM picks and answers chat messages as they arrive, prioritizing direct questions and supporters.
Streaming text-to-speech with lip sync, and Live2D expressions that follow the mood of each reply.
A written character profile guides each response. We are testing persona consistency and resistance to instructions embedded in viewer messages.
A desktop studio with a stop button, an OBS overlay, and per-session cost and latency reports.
Our first product is a close-up VTuber for Korean live streaming on CHZZK. The Rust/Tauri desktop app and Vue studio are in active development. The local server, chat replay, streaming speech, Live2D expressions, OBS overlay, stop control, and session reports have been exercised in development tests. A live creator pilot is the next milestone.
The desktop shell runs the local server and studio. Closing a session saves its turn and usage reports. Live CHZZK broadcast validation remains open.
A five-minute replay with real ElevenLabs speech recorded a 1.30-second median from chat to first audio; the 90th percentile was 5.82 seconds. This was a controlled development run in Germany, not a customer production benchmark.
We compared 14 model candidates across 40 prompts and three repeats. We use these tools to measure dialogue quality, latency, and cost before selecting a model.
Development results recorded October 3, 2026. We are preparing for private creator testing and have not raised external investment.
Claude Code is already part of our development workflow for the desktop app, streaming pipeline, and tests. The dialogue prototype currently uses other model providers. Our next step is to evaluate the direct Claude API for character dialogue and model-assisted evaluation, with a focus on persona consistency, prompt-injection resistance, and useful responses to Korean chat.
These are planned milestones. Claude is not yet the deployed dialogue model.
I am the developer behind MCP-Atlassian, an open-source server that connects AI assistants, including Claude, to Jira and Confluence. That work spans MCP tools, OAuth, multi-user access, and ongoing community maintenance.
Public MCP-Atlassian repository counts, checked October 8, 2026. These describe my open-source track record, not Animalive customers or revenue.
Soomiles is the name of my bootstrapped, unincorporated venture in Seongnam, South Korea. Animalive is its current product project, developed with a creative collaborator. We began this venture in 2026; the VTuber implementation has repository history from October 2026. The software is in private alpha.