Omnigram

About Omnigram

Jellyfin for videos. Immich for photos. Omnigram for books.

Omnigram is an AI-native, self-hosted book library and reading service. Deploy it on your NAS or homeserver, and your ebook collection becomes an intelligent, personal reading experience — not just a pile of files.

Origin

I'm a tech enthusiast with a NAS full of PDFs and EPUBs. Over time, files became messy, searching became painful, and the books I'd read faded from memory. No existing tool solves both "library management" and "AI-enhanced reading" — so Omnigram was born.

The Gap

                    Server (NAS/Self-hosted)       No Server (Client only)
                 ┌──────────────────────┬──────────────────────┐
   AI-capable    │                      │  Anx Reader          │
                 │    ❌ EMPTY!         │  Readwise (SaaS)     │
                 │                      │                      │
                 ├──────────────────────┼──────────────────────┤
   No AI         │  Calibre-Web         │  KOReader            │
                 │  Kavita              │  Foliate             │
                 │  Audiobookshelf      │                      │
                 └──────────────────────┴──────────────────────┘

"Self-hosted + AI + Reading" = zero competition. That's where Omnigram lives.

Design Philosophy

AI is Air, Not a Button

This is the fundamental difference between Omnigram and every competitor.

Most "AI readers" slap a ChatGPT dialog next to the book viewer. That's not AI reading — that's reading + chatting.

Omnigram's AI is invisible:

Turn to a new chapter, a context bar fades in with a recap — you didn't press anything
Select a word, a tooltip appears with a definition — you didn't "open AI"
Your bookshelf groups by topic automatically, imported books get tags and summaries
Your notes aren't scattered highlights — they're a knowledge network organized by concept

You're not "using AI." You're reading — and the experience itself has changed.

Companion, Not Tool

Reading has always been a solitary activity. AI can change that.

Omnigram's AI is a reading companion with configurable personality — like TARS from Interstellar, you can adjust its proactivity, style, depth, and warmth. It can be a silent assistant, a Socratic mentor, or a friend reading alongside you.

Insight, Not Statistics

Traditional reading apps give you bar charts and heatmaps — "you read 300 pages this month." So what?

Omnigram's Insights page tells the story of your reading journey: what domains you've been exploring, which books connect to each other, where your reading is heading. Not data — self-awareness.

Privacy as Freedom

Some books you'd rather keep private. Omnigram has a fully isolated "hidden library" — zero trace in the main interface, biometric entry, complete AI experience inside. Privacy protects against others' eyes, not your own reading experience.

Features

Reading Experience

Multi-format ebook reading (EPUB, MOBI, AZW3, FB2, TXT)
iOS, Android, macOS, Windows native client
Four-tab navigation: Reading Desk | Library | Insights | Settings
Reading Desk — open the app, see what you're reading, continue in one tap
Full-screen immersive reader with customizable typography
Reading notes, highlights, and bookmarks
TTS text-to-speech with customizable voices

AI-Powered (Ambient)

AI reading companion with configurable personality (TARS model — adjust proactivity, style, depth, warmth)
Context bar — chapter recaps fade in automatically when you turn pages
Inline glossary — select a word, see AI-generated definitions without leaving the page
Margin notes — AI surfaces cross-book connections in the page margin
Smart bookshelf — auto-tagging, summaries, and reading time estimates on import
Semantic search — search your library by meaning, not just keywords
Knowledge network — notes organized by concept across all your books
Reading narrative — AI tells the story of your reading journey, not just charts

Server

Self-hosted book library with directory scanning
Multi-format metadata extraction
Multi-user management with session auth
Book search, favorites, reading progress sync across devices
Docker one-click deployment (SQLite / PostgreSQL / MySQL)
Server-side TTS via sidecar (Kokoro / Edge TTS, OpenAI-compatible API)
OPDS catalog protocol

Privacy

Hidden private library — completely isolated second space with biometric access
Zero trace in main interface — private books never appear in bookshelf, search, or insights
Full AI experience inside private space
Encrypted local storage with platform keystore

Data Portability

Export notes/highlights as Markdown, JSON, or CSV
Export knowledge network for Obsidian/Logseq
Import highlights from Kindle, Apple Books, Readwise
Full library export as portable archive
OPDS catalog for interoperability

An Analogy

If Calibre-Web is a library bookshelf — you come find a book, you read it yourself — then Omnigram gives you a personal librarian.

They remember what you've read, what you like, where you left off. Occasionally they'll say, "You might like this one — it connects to that book you read last month." They don't talk much, but when you need them, they know everything.

Architecture

graph TB
    subgraph Client["📱 Omnigram App (Flutter)"]
        Desk["📖 Reading Desk"]
        Library["📚 Library"]
        Insights["💡 Insights"]
        Settings["⚙ Settings"]
        Reader["📄 Immersive Reader<br/>(EPUB / foliate-js)"]
        Companion["🤖 AI Companion<br/>(TARS personality)"]
        LocalDB["💾 sqflite"]
    end

    subgraph Server["🖥️ Omnigram Server (Go + Gin)"]
        API["REST API<br/>(Auth · Reader · Sync · AI · TTS · OPDS)"]
        BookMgr["📚 Book Manager<br/>(scan · metadata · import)"]
        AIService["🧠 AI Service<br/>(summary · tags · search)"]
        SyncService["🔄 Sync Service<br/>(progress · notes · WebDAV)"]
        DB["🗄️ SQLite / PostgreSQL"]
        Store["📦 BadgerDB<br/>(key-value store)"]
    end

    subgraph External["☁️ External Services"]
        LLM["LLM Provider<br/>(OpenAI · Claude · Gemini<br/>Ollama · DeepSeek)"]
        TTS["TTS Sidecar<br/>(Kokoro / Edge TTS)"]
    end

    subgraph NAS["🏠 NAS / Homeserver"]
        Docker["🐳 Docker Compose"]
        Storage["📁 Book Storage<br/>(/books volume)"]
    end

    Desk --> Reader
    Library --> Reader
    Reader --> Companion
    Client -->|REST API| API
    Client -->|WebDAV| SyncService

    API --> BookMgr
    API --> AIService
    API --> SyncService
    BookMgr --> DB
    BookMgr --> Storage
    AIService --> LLM
    SyncService --> DB
    API --> Store

    Docker --> Server
    Docker --> TTS
    Server --> TTS

Official Documentation

You can find the official documentation (including installation manuals) at https://omnigram.lxpio.com/.

Examples

For the mobile app, you can use https://omnigram-demo.lxpio.com:9443 for the Server Endpoint URL

Credentials:
email: admin
password: 123456

For Dev

Build

Omnigram App (Flutter)

git clone https://github.com/lxpio/omnigram.git
cd omnigram/app
flutter pub get
flutter gen-l10n
dart run build_runner build --delete-conflicting-outputs
flutter build apk  # or: flutter build ios / macos / windows

Omnigram Server (Go)

git clone https://github.com/lxpio/omnigram.git
cd omnigram/server
make

# Docker build
make docker

TTS Service

Omnigram Server supports TTS via a sidecar Docker container with OpenAI-compatible API. The default choice is Kokoro-FastAPI:

# docker-compose.yml — add TTS sidecar
services:
  tts:
    image: ghcr.io/remsky/kokoro-fastapi:latest
    # OpenAI-compatible API: POST /v1/audio/speech
    # CPU mode by default, add GPU resources for acceleration

Configure TTS_PROVIDER=kokoro and TTS_SIDECAR_URL=http://tts:8880 in your Omnigram Server environment.

Edge TTS is also available as a zero-deployment fallback (no additional container needed).

Tech Stack

Component	Technology
Server	Go 1.23 + Gin + GORM + BadgerDB
Client	Flutter 3.41 + Riverpod v2
AI	langchain_dart (OpenAI, Claude, Gemini, Ollama, DeepSeek)
TTS	Kokoro / Edge TTS (server sidecar) + sherpa-onnx (on-device)
Database	SQLite / PostgreSQL / MySQL
Deployment	Docker / Docker Compose

Acknowledgments

Omnigram's reading engine builds on Anx Reader (MIT License). We are deeply grateful for their excellent work.

Key libraries and dependencies:

Anx Reader — Reading engine foundation
foliate-js — Ebook rendering engine
Riverpod — State management
langchain_dart — AI integration
Kokoro-FastAPI — TTS engine (Sidecar)

License

This project is licensed under the MIT License - see the LICENSE.md file for details

omnigram