Infinite Bookshelf 开源：输入主题即可流式生成整本结构化书籍

Turn a topic into a complete, structured book, written live by the AI model of your choice.

Bring your own API key · 16 providers and any OpenAI-compatible API · Self-hosted in one container

Try it live at ib.sasivarnasarma.me →

Quick start · Features · How it works · Privacy · Docs

✨ What it is

Type a topic, such as "The history of tea" or "Linear algebra for programmers". Infinite Bookshelf drafts a table of contents, lets you edit it, then writes the book section by section. You watch every word stream in, the way a model writes it.

Each section knows the outline and what earlier sections already covered, so chapters build on each other instead of repeating. You choose the model for every step, can switch it at any time, and can rewrite any section with a note. When the book is done, export it as Markdown, a typeset PDF, or a backup file.

It runs on your API keys, which never leave your browser except to make a request, and your books never leave your browser at all.

🚀 Quick start

Tip

Just want to try it? Open ib.sasivarnasarma.me and add a key from any provider: nothing to install. Your key and book text pass through that server for each request (never stored or logged); run your own copy if you'd rather they didn't.

🐳 Docker (recommended)

git clone https://github.com/Sasivarnasarma/infinite-bookshelf.git
cd infinite-bookshelf
docker compose up -d

Open http://localhost:9752.

💻 From source

Needs uv, Node.js 22+ and pnpm.

pnpm bootstrap         # install everything
cp .env.example .env   # optional settings
pnpm dev               # API on :8000, web app on :5173

Open http://localhost:5173.

📦 Without Docker

To run the finished app on a server without Docker, build it once and start it:

pnpm bootstrap         # install everything
pnpm build             # build the web app
pnpm start             # serve the app and the API → http://localhost:9752

One process serves everything on port 9752, like the container. See Self-hosting → Without Docker for keeping it running with systemd.

Then go to Settings → Providers & models, add a provider, and paste your API key. Google Gemini has a generous free tier, and Ollama runs models on your own machine with no key at all.

Tip

On GitHub Codespaces, or anywhere the container can't look up domain names, run docker compose up -d infinite-bookshelf-host instead. It uses the host's network. See Self-hosting.

🌟 Features

📝 Write

Outline first. Review the drafted contents before anything is written: rename, reorder by drag and drop, change levels, add or remove entries.
Live streaming. Every section appears word by word as the model writes it.
Connected chapters. Each section sees the outline and a digest of earlier sections.
Whole book or chapter by chapter. Stop after each chapter to read it before the next.
Pause and resume at any time, even after closing the tab.
Your direction: section length, writing style, depth, guidelines, and your own notes or a .txt/.md file as source material.

🤖 Models

16 built-in providers, plus any OpenAI-compatible API.
A model per step: outline, title and chapters can each use a different model, even from different providers.
Switch models whenever the book is paused; each section records which model wrote it.
Several keys per provider, with automatic switching when one fails and optional rotation.
Suggestions of the best, balanced and fast models you have for each step.
Local models through Ollama and LM Studio.

🪄 Refine

Rewrite any section with a note, such as "add a worked example", using the same model or another one. The original stays until the new version is finished.
Maths ($a^2 + b^2 = c^2$), tables and highlighted code, in the reader and the PDF.
Follow along: the reader scrolls with the section being written, until you scroll away.

📦 Export and more

Markdown, a typeset PDF (title page, contents with page numbers, maths), or a JSON backup to move books between devices.
Light and dark themes, three reading sizes.
Works on phones and tablets, and installs to the home screen like an app.
Link previews with an image when you share it on WhatsApp, Slack, Discord, X and more.
Interactive API docs at /api/docs.

Supported providers

Kind	Providers
Model makers	OpenAI · Anthropic Claude · Google Gemini · xAI Grok · DeepSeek · Mistral AI · Moonshot Kimi · Alibaba Qwen · Z.ai GLM
Gateways and fast inference	OpenRouter · Groq · Together AI · Fireworks AI · Cerebras
On your own machine	Ollama · LM Studio
Anything else	Any OpenAI-compatible API, added with its base URL (vLLM, LiteLLM, llama.cpp server…)

Free tiers: Google Gemini, Groq and Cerebras. Ollama and LM Studio need no key.

🔍 How it works

flowchart LR
    subgraph Browser["🌐 Your browser"]
        UI["Web app<br/>(React)"]
        Keys[("API keys<br/>localStorage")]
        Books[("Books<br/>IndexedDB")]
    end
    subgraph Server["🖥️ Infinite Bookshelf server (stateless)"]
        API["FastAPI<br/>prompts · streaming · PDF"]
    end
    Provider["☁️ AI provider<br/>(OpenAI, Gemini, Ollama…)"]

    Keys -. key, per request .-> UI
    UI <--> Books
    UI -- "topic, outline, book so far + key" --> API
    API -- "model request" --> Provider
    Provider -- "tokens" --> API
    API -- "Server-Sent Events" --> UI

Outline. The server asks your outline model for a table of contents and your title model for a title, streaming both back.
Review. You edit the outline in the browser, or skip review and start writing at once.
Write. The browser asks for one section at a time, sending the outline and the sections written so far. The server builds the prompt and streams the model's words back. Each finished section is saved in your browser straight away.
Export. Markdown and backups are made in the browser; the server typesets PDFs.

The server stores nothing: no database, no accounts, no sessions. Read the architecture guide for the details.

🔒 Your data stays yours

	Where it lives	What the server does
API keys	Your browser (or only this tab)	Uses the key for one request, then forgets it. Never stored or logged.
Books	Your browser's IndexedDB	Receives the outline and text needed for each request, then forgets them.
Settings	Your browser	Nothing.

The page loads no third-party scripts, fonts or trackers, under a strict Content Security Policy.
Turn off Remember my keys and they're kept only until you close the tab.
Back books up from Settings → Your data; they exist nowhere else.

Important

Running a server for other people? Read Running a public instance first. The default Docker Compose file is set up for a private server.

📚 Documentation

Guide	What's in it
Self-hosting	Docker, configuration, local models, reverse proxies, public instances, troubleshooting
Architecture	How the browser and server work together, streaming, prompts, PDF export, security
API reference	Endpoints, request bodies, streamed events, error codes
Web app	The web app's commands, `IB_API_URL`, and conventions
Contributing	Development setup, checks, tests and commit style
Security policy	How to report a vulnerability, and what's protected
Changelog	What changed in each release

📁 Project layout

infinite-bookshelf/
├── api/                  Python API: FastAPI, uv, pytest
│   ├── src/infinite_bookshelf/
│   │   ├── engine/       book-writing logic: prompts, providers, PDF (no web code)
│   │   └── server/       HTTP layer: routes, streaming, security, settings
│   └── tests/
├── web/                  Web app: React 19, TypeScript, Vite, Tailwind CSS, Vitest
│   └── src/              pages, components, and lib/ (the writing loop, storage, settings)
├── docs/                 Architecture and self-hosting guides
├── Dockerfile            One image: the API serving the built web app
├── docker-compose.yml    Self-hosting on port 9752
└── .env.example          Every setting, with its default

Command	What it does
`pnpm bootstrap`	Install the web and API dependencies
`pnpm dev`	Run the API (`:8000`) and web app (`:5173`), both reloading on changes
`pnpm check`	Everything CI runs: formatting, lint, typecheck and all tests
`pnpm test`	Web tests (Vitest), then API tests (pytest)
`pnpm format`	Format the web app and docs (Prettier) and the API (Ruff)
`pnpm build`	Production build of the web app
`pnpm start`	Serve the built web app and the API on `:9752`, without Docker

🤝 Contributing

Issues and pull requests are welcome. Start with CONTRIBUTING.md; please report security problems privately, as described in SECURITY.md.

💛 Credits

Infinite Bookshelf began as the original Infinite Bookshelf by Benjamin Klieger, and this project is built on that idea.

_{Provider names and logos are trademarks of their owners, shown only to indicate compatibility.}

📄 Licence

Distributed under the MIT License