notegenius

The open-source NotebookLM alternative
Upload sources, chat with your documents, generate insights. Local document storage; cloud AI through Anthropic. Provider usage limits and charges apply.

Upload PDFs, paste plain text, add web article URLs, or drop in a YouTube link — transcripts are fetched automatically. All content is chunked and indexed in SQLite FTS5 for fast retrieval.

Ask anything about your documents. Every answer cites its sources with [1] markers. Click any citation to see the exact passage from the original document in a popover — not just the source name.

All generations stream in real-time and are saved to your notebook.

Write Markdown notes alongside your research. Auto-saves as you type, supports pinning, and keeps everything per-notebook.

flowchart TD
    subgraph Sources["📂 Source Ingestion"]
        PDF["PDF\n(pdf-parse)"]
        URL["Web URL\n(cheerio + axios)"]
        YT["YouTube\n(transcript API)"]
        TXT["Plain Text"]
    end

    subgraph Processing["⚙️ Processing Pipeline"]
        Chunker["Sliding-window\nChunker\n1200 chars / 200 overlap"]
        SQLite[("SQLite\n+ FTS5 Index")]
    end

    subgraph Retrieval["🔍 Retrieval"]
        FTS["BM25 Full-text\nSearch (FTS5)"]
        TopK["Top-K Chunks\n(default: 8)"]
    end

    subgraph AI["🤖 Claude AI"]
        Chat["Streaming Chat\nclaude-sonnet-4-6"]
        Generate["Generation\n7 types"]
        Citations["Citation\nExtraction"]
    end

    subgraph UI["🖥️ Next.js Frontend"]
        Sources2["Sources\nPanel"]
        ChatUI["Chat\nPanel"]
        GenerateUI["Generate\nPanel"]
        Notes["Notes\nPanel"]
    end

    PDF & URL & YT & TXT --> Chunker
    Chunker --> SQLite
    SQLite --> FTS
    FTS --> TopK
    TopK --> Chat
    TopK --> Generate
    Chat --> Citations
    Citations --> ChatUI
    Generate --> GenerateUI
    Sources2 --> SQLite
    Notes --> SQLite

sequenceDiagram
    actor User
    participant UI as Browser
    participant API as Next.js API
    participant DB as SQLite
    participant Claude as Claude API

    User->>UI: Upload PDF / paste URL
    UI->>API: POST /api/sources (SSE)
    API->>DB: Create source row
    API-->>UI: SSE: status=processing
    API->>DB: Insert chunks + FTS index
    API-->>UI: SSE: status=ready
    UI->>UI: Show ✅ in Sources panel

    User->>UI: Ask a question
    UI->>API: POST /api/chat (SSE)
    API->>DB: FTS5 BM25 search → top-K chunks
    API->>Claude: Stream with source context
    Claude-->>API: Stream tokens + [1][2] citations
    API-->>UI: SSE: text chunks
    UI->>UI: Render message + citation popovers

Frontend          Backend           AI              Data
─────────         ────────          ──              ────
Next.js 16        Next.js API       Claude          SQLite
TypeScript        Routes            Sonnet 4.6      better-sqlite3
Tailwind v4       SSE streaming     @anthropic-ai   FTS5 BM25
Radix UI          better-sqlite3    /sdk            Sliding-window
Sonner toasts     pdf-parse                         chunking
next-themes       cheerio
                  youtube-transcript

• Node.js 18+
• A Claude API key → console.anthropic.com

git clone https://github.com/ArttuAn/notegenius
cd notegenius
npm install

cp .env.example .env.local

Edit .env.local:

CLAUDE_API_KEY=sk-ant-your-key-here

npm run dev

Open http://localhost:3000. The SQLite database is created automatically at ./data/notegenius.db on first run.

notegenius/
├── app/
│   ├── api/
│   │   ├── chat/          # Streaming Q&A + session management
│   │   ├── generate/      # 7-type AI generation (SSE)
│   │   ├── notebooks/     # CRUD for notebooks
│   │   ├── notes/         # CRUD for notes
│   │   ├── settings/      # App settings
│   │   └── sources/       # Source ingestion (SSE)
│   ├── notebook/[id]/     # Workspace page
│   └── notebooks/         # Dashboard page
│
├── lib/
│   ├── ai/
│   │   ├── chat.ts        # FTS5 retrieval → Claude streaming + citation parsing
│   │   ├── generate.ts    # 7 generation types → Claude streaming
│   │   └── prompts.ts     # All system prompts
│   ├── db/
│   │   ├── schema.ts      # SQLite schema + TypeScript types
│   │   ├── migrate.ts     # Idempotent migrations
│   │   └── queries/       # Typed query helpers per entity
│   └── ingestion/
│       ├── chunker.ts     # Sliding-window text chunker
│       ├── pdf.ts         # pdf-parse v2 extraction
│       ├── url.ts         # cheerio web scraper
│       └── youtube.ts     # Transcript fetcher
│
└── components/
    ├── notebook/
    │   ├── ChatPanel.tsx      # Streaming chat + citation chips
    │   ├── SourcesPanel.tsx   # Source list + upload dialog
    │   ├── GeneratePanel.tsx  # Generation type selector + results
    │   └── NotesPanel.tsx     # Auto-saving markdown notes
    └── ui/                    # Radix UI primitives

MIT — use it, fork it, ship it.

Set NOTEGENIUS_PASSWORD to a unique random value of at least 32 characters before starting. The browser prompts for username admin (override with NOTEGENIUS_USERNAME) and that password. All pages and API handlers require the same owner credentials. This is a single-owner app: everyone given those credentials has access to the same notebooks. It does not support separate user accounts or tenant isolation. Missing credentials fail closed.

Loopback origins are allowed by default. For a deliberate remote deployment, set NOTEGENIUS_ORIGIN to the exact public origin, without a trailing slash, and use HTTPS. Other hosts and browser origins are rejected. URL imports allow only public HTTP(S) targets on standard ports, validate every redirect, pin the verified DNS result to the connection, and cap response size and duration.

Documents are stored in local SQLite. Chat and generation send selected source excerpts, prompts, and conversation context to Anthropic. URL/YouTube import also contacts external source providers. Do not use confidential data unless you are comfortable with those transfers and the provider's data policy. Provider spending caps are a separate deployment control. Development and tests now require Node 22.12 or newer.