19FM: Solana 24/7 Auto Streaming Station

📡 19FM Architecture Update — Going Audio-First Big direction shift. We will remove some server-side VRM rendering and HLS video streaming, and rebuilding around audio. The AI host still reacts to Solana trades with real voice. Comments, chat, /ask — all staying. Why we're doing this and what comes next 🧵👇

This is a direction, not a release. We'll share progress as we build. The core promise stays the same: tokens, AITubers, and people — connected. 🔗 Current demo still live: 19fm.app/ch/19fm #BuildInPublic (8/8)

📡 19FM Architecture Update — Going Audio-First Big direction shift. We will remove some server-side VRM rendering and HLS video streaming, and rebuilding around audio. The AI host still reacts to Solana trades with real voice. Comments, chat, /ask — all staying. Why we're doing this and what comes next 🧵👇

What the experience looks like after: You open a station. Danmaku flows across the screen. The AI speaks and a waveform pulses. Price ticker updates. Chat is live. /ask works. The vibe — a place to hang out around a token — doesn't change. It just gets lighter and scales. (7/8)

What this unlocks: ✅ 100+ stations without GPU cost ✅ On-demand station spin-up (no manual infra) ✅ Monthly cost stays ~$100 regardless of station count ✅ Works on any phone/laptop — no heavy rendering ✅ Clean separation for multi-project operation 1 station used to cost a full GPU pipeline. Now it costs a row in a database. (6/8)

New architecture: Solana TX → PartyKit (event hub) → viewer's browser The browser handles everything: 🎵 Audio playback (AI voice via TTS) 📊 Danmaku overlays (CSS/Canvas, zero GPU) 🎤 Audio visualizer (waveform reacts to AI speech) 📝 Subtitles of what the AI is saying 💬 Chat + /ask Server only does: TX monitoring, LLM calls, TTS generation. That's it. (5/8)

The answer: the AUDIO is the product. 19FM's value is "an AI host that watches trades and talks about them." You don't need to stare at a 3D avatar for that. You need to hear it. Most users will have 19FM open in a background tab anyway — like radio. (4/8)

3/8 We want to support 100+ token stations. Under the current architecture, that means 100x the GPU load and 100x the cost. GPU servers for VRM rendering at scale = 10-100x our budget. It simply doesn't work. So we asked: what's actually essential?

The problem with our current stack: Right now, ONE station requires: → LLM inference → TTS synthesis → VRM 3D rendering → ffmpeg video encoding → HLS streaming via nginx All on a single local machine. If it goes down, everything goes down. And adding a second station means doubling all of that.

The Build in Public Hackathon application deadline is being extended by 1 week! Due to significantly more interest from teams looking to launch their project & apply, the Pump fun BiP Hackathon application deadline has been extended to February 25, 2026. Learn more 👇

その他の成果物: ✅ 共有ステーション（/ch/shared）の試験実装 ✅ TTS フェイルセーフ経路を追加（片方のエンジンが止まっても合成が死なない） ✅ 共有ステーションのインフラ改善・安定化 次: かんたん AITuber 作成＆申請フォーム・審査周りの改善

🚢 19FM Daily Ship (Hackathon Eve Day?) Today's major ship: Bilingual AI 🎙️ AI host now auto-detects language and switches voice engines on the fly. - Japanese comments → TTS A - English comments → TTS B Now AI talks back in your language. Try it 👇 19fm.app/ch/19fm

Also shipped: ✅ Shared Station (/ch/shared) (Experimental) PoC ✅ TTS fail-safe path added — if one engine hiccups, synthesis doesn't die ✅ Shared-station infrastructure update Next up: AI host submission form enhancement

🇯🇵 日本語まとめ 今日の成果物: AIホストが日英バイリンガルになりました。 コメントの言語を自動判定し、音声合成エンジンを切り替えます。 日本語で話しかければ日本語で、英語なら英語で返ってきます。 （声が統一感ないので、これは要調整です） 19fm.app/ch/19fm

Built New Article about 19FM. We participates @Pumpfun Build in Public Hackathon #19FM

19FM の実験的機能として、「だれでも AI VTuber 即反映」の共有ステーション機能を作っています。 - 1個の共有ステーションで固定 - AI VTuber 即作成UI - クイック審査して問題なければ共有ステーションに即反映 - 1人ずつローテ できれば明日までに出したい #19FM

Now i'm implementing "Shared Station with Quick AI VTuber Creation". In short, - Fixed one shared station - Everyone can create AI VTuber in a few minutes - Now your AI VTuber is Live (rotation) Very experimental concept but it is good for hackathon

Any request ? Feel free to reply or comment on our LIVE website. リクエストがあれば気軽にリプライください！

3. Entertainment features 🎨 Music Genre backgrounds Added genre-specific background images so viewers can tell the active genre at a glance. ジャンル別の背景画像を追加し、再生中のジャンルの雰囲気にあった画像が流れるようにしました。

📻️19FM Ship Log Today 1. AI VTuber is now live — an AI radio DJ powered by LLM + TTS + VRM avatar. She reacts to viewer comments in real-time. 実際に会話が出来る AI VTuber の試験版を稼働開始。LLM + TTS + VRM アバターで実装。視聴者のコメントにリアルタイムで反応します。