iArt.ai
AI 모션 그래픽 소프트웨어, 누구나 사용할 수 있는 After Effects 대안, 텍스트 프롬프트로 몇 분 안에 애니메이션·설명 영상 제작
개인 계정의 공개 저장소 소유자는 GitHub로 로그인해 운영자를 확인하고 소개를 직접 관리할 수 있습니다. 관리자 승인은 필요하지 않습니다.
로그인한 뒤 확인 파일을 만들어 이 저장소의 기본 브랜치에 30분 안에 커밋해 주세요. 다른 브랜치나 포크의 파일은 인정하지 않습니다. 파일을 올린 뒤 아래에서 확인하면 관리자 승인 없이 관리할 수 있습니다. 자격을 유지하려면 파일을 남겨 두세요.
로그인이 필요하면 로그인 화면으로 이동합니다. 로그인 후 이곳으로 돌아와 원하는 확인 버튼을 다시 눌러 주세요.
프로젝트 폴더에서 /nomorevibe verify를 실행하면 소유권을 확인하고 소개를 직접 관리할 수 있습니다.
1분 만에 확인하기A quiet place to write a long story with a language model. Chapters, not one endless chat — talks to any OpenAI-compatible endpoint, runs on your machine, keeps your stories as plain JSON.
저장소 README에서 가져옴
README 전체를 불러오는 중입니다. 아래는 저장된 발췌입니다.
A quiet place to write a long story with a language model.
Runs on your own machine, talks to any OpenAI-compatible endpoint, and keeps your work as
plain JSON files you can read without it.
Download it for Windows or Linux →
No Node.js, no terminal. macOS is not built; that page says why.
Lamplit is a single-page app for collaborative fiction: you write a line, a model writes
the next passage, and the two of you keep going. It is built around one idea that most tools of
this kind leave out — a story is written in chapters.README 전문 보기 ↗Lamplit is a single-page app for collaborative fiction: you write a line, a model writes the next passage, and the two of you keep going. It is built around one idea that most tools of this kind leave out — a story is written in chapters.
Each chapter opens on a scene you write yourself, runs for as long as it wants to, and is then closed: the model folds it into a running summary of everything that has happened, and the next chapter starts from there. Only the current chapter is ever sent to the model, so a story that runs for months still costs the same per reply as one that started this morning.
There is no account, no cloud, and no server between you and the model. The browser talks to your endpoint directly. A small local server sits behind the app to hold your stories — plain JSON files, and the only copy there is.
It is deliberately not a configuration tool. No character cards, no extensions, no prompt manager, no images. One bar across the top, and everything in it is something you will actually use. If you want to know exactly what will be sent, one click shows you the whole assembled prompt, block by block.
Tools for this are usually built for people who enjoy tuning them. This one is built for people who want to read what comes out. Two things follow from that:
| Chapters with scenes | A chapter cannot be written into until its scene is written — the one thing the app insists on. The scene goes to the model verbatim and stays editable. |
| Close a chapter | The model is handed the story so far and asked to give the whole thing back with this chapter folded in. You edit the result before it lands. |
| A world that fires on keywords | Facts, people and places that are only sent when the story mentions them, plus a "story so far" that is always sent. |
| Narrator or role-play | One omniscient voice, or a named cast the model plays while never writing for you. |
| Book-style reading | Markdown, quoted speech set apart, *actions* in italics, adjustable text size, light and dark. |
| Per-message control | Edit, regenerate, replay from here, copy, delete. Stop mid-stream keeps the partial answer. |
| Every provider on the list, or your own | OpenAI, Anthropic, Google, OpenRouter, NanoGPT, Ollama and the rest, each filling in its own URL and linking to where it hands out keys — or paste any URL that answers /models and /chat/completions. Streaming, with the provider's real token usage shown after each reply. |
| Plain files on disk | settings.json, stories/<id>.json, chapters/<id>.json — read once when the app starts and written straight back, with no second copy in the browser to fall out of step with them. Copy the folder and you have copied everything. Zipped to backups/ once a day. |
| Every chapter opens on a scene | Nothing is hidden |
| A world that only speaks when spoken to | Close a chapter and carry the story forward |
Nothing below is needed to use Lamplit — the download page has installers that carry Node inside them. This is the way in for people who want the source.
Node 22.22+, 24.15+ or 26+ — what Angular 22 asks for.
npm install
npm start
Then open http://localhost:4200. The app asks for a model first, then who tells the story and who you play, then the opening scene — and you are writing.
npm run package
builds a self-contained folder and a ~1 MB zip — the same one every release publishes as
Lamplit.zip.
Unzip it anywhere, run start.bat (Windows), start.command (macOS) or ./start.sh (Linux), and
the app opens in your browser with no install step at all.
See Running it anywhere.
npm run desktop:dist
builds the desktop installers for whatever OS you are on. See The desktop app.
| Getting started | Installing, the first run, writing your first chapter |
| Reading and writing | The page, the composer, what each message can do |
| Chapters | Scenes, closing a chapter, the chapters list |
| Story and world | Narrator or role-play, your persona, the world and its lore |
| The prompt | How every request is assembled, and how to look at it |
| Models and parameters | Connecting, picking a model, sampling settings |
| Your data | Where files live, backups, offline, several stories |
| The desktop app | One download, one icon, and where it keeps your stories |
| Running it anywhere | Building the zip and running it on another machine |
| Development | Repo layout, scripts, tests, how the pictures are made |
Your key is stored in plain text, in data/settings.json on your own machine. That is deliberate:
Lamplit is a single-user tool on your own computer, and a local file you control beats a
secret store you have to unlock every time. The server listens on 127.0.0.1 only — nothing else
on your network can reach it — unless you turn sharing on under Preferences → Advanced, which
opens a second listener your phone can reach after it has scanned a code. A phone that has scanned
it can read the key, so share on a network you trust. Don't run Lamplit on a machine you share, and
don't put it on the open internet.
Written in four steps, all done: streaming chat and the model connection, then chapters and the
world, then persistence and packaging, then a desktop build and a way for people to find it.
PLAN.md is the plan of record and records why each decision went the way it did.
Windows and Linux builds are not signed — Windows will warn about an unfamiliar app the first time, and the download page shows the two clicks past it. macOS is not built at all: a build that opens without a fight needs Apple's developer licence, which this project does not hold. If you have one and would like to contribute those builds, an issue would be very welcome.
Not here, on purpose: images, group chats, swipes and variants, an extensions system, text completion (non-chat) endpoints.
Angular 22 (standalone, signals, zoneless) · Angular Material · Express 5 · Playwright · Vitest.
No state library, no HTTP client, no model SDK — fetch and a hand-written SSE reader.
MIT. Do what you like with it.
The book in the icon is Lucide’s book-open-text, under the ISC licence — see NOTICE.