═══════════════════════════════════════
Teach AI How to Act
You teach it how to do — and it tries.
═══════════════════════════════════════
[/p]A little desktop toy where you teach an AI how to do something — just by typing.
You type out a thing — how to take a shower, which hand goes in first — and the AI takes a guess, tries to figure out what you meant, and shows you what it came up with. Whether it got it right is up to you to judge.
It doesn't actually learn anything. It just tries to understand what you wrote, and then it shows you how much it understood. You decide if it did it right.
🎬 What's the deal
📝 You say one line · It takes one shot
You type "first reach with the left hand" / "now raise the right" — the AI reads the line, flips through whatever it knows, and picks something to play back. You watch and decide: did that match what I meant? If not, keep teaching.}
🖊️ You teach — type, send, tell it what to do
🧠 It tries — reads your words, digs through its own library, picks something
🎞️ You watch — it shows you what it found, you judge if it's right
🎥 How to see what it picked
Full-screen video is how it shows you the result — no menus, no controls, just the clip it chose. You watch and see if it matches what you tried to teach. The chat drawer sits quietly in the corner until you need it.}
🎞️ Seamless cuts — segments preload so there's no blackout or hiccup
🔲 Doesn't block — the teach drawer stays out of the way until you expand it
🖱️ Click and teach — pick a series from the home page, start teaching right away
🤖 Local model · fully offline
Everything runs on your own machine. No internet needed, no account, no uploads. The local LLM (a GGUF model you can replace) does the guessing right here.}
⚡ Plug and play — first launch wakes the local model automatically
🔌 Swappable — drop any GGUF model you like into the models folder
🛡️ Privacy — every teach goes nowhere but your machine
🖥️ Minimal interface
Two states: pick a series → watch full-screen + type in the teach drawer. That's it. Nothing else gets in the way.}
🖱️ Click a card to enter a series, click the teach drawer to type
⌨️ F11 fullscreen / Esc back to home / Enter to open the teach drawer
🎮 Legacy W/D/S/A frame controls are still there if you need frame-by-frame inspection
🔧 Tech stack glance
▼ 01｜User layer
Series picker · Full-screen player · Teach drawer · Queue HUD · Context menu
▼ 02｜AI teach core
AIDirector · Local LLM · Instruction parse · Action match · Fallback chain
▼ 03｜Show engine
Dual-video pre-swap · No-gap transition · Buffer preload · Event sync
▼ 04｜State layer
Play queue · Current index · History cache · Two-way sync
▼ 05｜Local AI runtime
llama-server · GGUF model · Port 9141 · Auto start/cleanup
▼ 06｜Data layer
Local series config · Local playlist · Video assets · Fully offline
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
This software includes modified Tauri 2 application, llama.cpp (MIT) by ggml-org, Hunyuan 7B by Tencent (see model card for license), and FFmpeg (LGPL) as external tool. Full license texts are included in the installation directory.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
You teach it how to do something.
It tries to understand you.
Then it shows you what it got.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
