[kit] Add PageIndex NotebookLM for vectorless tree-structured RAG

kits/assistant/pageindex-notebooklm/.env.example

@@ -0,0 +1,12 @@
+# Lamatic Project Settings
+# Get these from: studio.lamatic.ai → Settings → API Keys
+LAMATIC_API_KEY="YOUR_LAMATIC_API_KEY"
+LAMATIC_PROJECT_ID="YOUR_LAMATIC_PROJECT_ID"
+LAMATIC_API_URL="YOUR_LAMATIC_API_ENDPOINT"
+
+# Flow IDs
+# Get each Flow ID from: Lamatic Studio → Flow → three-dot menu → Copy ID
+FLOW_ID_UPLOAD="YOUR_UPLOAD_FLOW_ID"
+FLOW_ID_CHAT="YOUR_CHAT_FLOW_ID"
+FLOW_ID_LIST="YOUR_LIST_FLOW_ID"
+FLOW_ID_TREE="YOUR_TREE_FLOW_ID"

kits/assistant/pageindex-notebooklm/.gitignore

@@ -0,0 +1,49 @@
+# Dependencies
+node_modules/
+.pnp
+.pnp.js
+
+# Build outputs
+dist/
+build/
+.next/
+out/
+
+# Environment variables
+.env
+.env.local
+.env.development.local
+.env.test.local
+.env.production.local
+
+# Logs
+logs/
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+
+# OS files
+.DS_Store
+Thumbs.db
+
+# IDE / Editor
+.vscode/
+.idea/
+*.swp
+*.swo
+
+# TypeScript
+*.tsbuildinfo
+
+# Cache
+.cache/
+.parcel-cache/
+.turbo/
+
+# Testing
+coverage/
+
+# Misc
+*.pem

kits/assistant/pageindex-notebooklm/README.md

@@ -0,0 +1,249 @@
+# PageIndex NotebookLM — AgentKit
+
+Upload any PDF and chat with it using **vectorless, tree-structured RAG** — powered **end-to-end by Lamatic AI flows**.
+
+> **No vector database. No external Python server. No custom backend code.**
+> Just 4 Lamatic flows + a Next.js frontend that implements the full PageIndex pipeline — from PDF ingestion to tree-navigated question answering — entirely within Lamatic's orchestration layer.
+
+---
+
+## What Makes This Different
+
+Most RAG implementations require a vector database, an embedding model, a retrieval server, and often a separate Python backend. **This kit eliminates all of that.**
+
+The entire PageIndex pipeline — TOC detection, tree construction, page indexing, summary generation, tree-navigated search, and LLM answering — is implemented as **4 Lamatic AI flows** with zero external servers or Python code. The Next.js frontend communicates exclusively with Lamatic's flow execution API via the official `lamatic` SDK.
+
+### Key Highlights
+
+- **100% Lamatic-powered backend** — all document processing, indexing, retrieval, and answering logic lives inside Lamatic flows
+- **No vector DB** — uses a hierarchical tree index (built from the document's table of contents) instead of vector embeddings
+- **No external server** — no FastAPI, no Railway, no Python — the Lamatic flows handle everything
+- **No chunking** — sections are identified by their structural position in the document, not arbitrary text splits
+
+---
+
+## Architecture
+
+```text
+┌────────────────────────────────────────────────────┐
+│                  Next.js Frontend                  │
+│  ┌──────────┐ ┌──────────┐ ┌────────┐ ┌─────────┐ │
+│  │ Document │ │   Chat   │ │  Tree  │ │Document │ │
+│  │  Upload  │ │  Window  │ │ Viewer │ │  List   │ │
+│  └────┬─────┘ └────┬─────┘ └───┬────┘ └────┬────┘ │
+│       │             │           │            │      │
+│  ┌────┴─────────────┴───────────┴────────────┴───┐  │
+│  │         Server Actions (orchestrate.ts)       │  │
+│  └────────────────────┬──────────────────────────┘  │
+│                       │                             │
+│            Lamatic SDK (lamatic npm)                │
+└───────────────────────┼─────────────────────────────┘
+                        │
+           ┌────────────┴────────────┐
+           │    Lamatic AI Platform   │
+           │                          │
+           │  Flow 1: Upload + Index  │
+           │  Flow 2: Chat + Retrieve │
+           │  Flow 3: List Documents  │
+           │  Flow 4: Tree / Delete   │
+           │                          │
+           │     ┌──────────────┐     │
+           │     │   Supabase   │     │
+           │     │  (PostgreSQL) │     │
+           │     └──────────────┘     │
+           └──────────────────────────┘
+```
+
+---
+
+## How It Works
+
+### Document Ingestion (Flow 1)
+
+When a PDF is uploaded, the Lamatic flow runs a multi-stage pipeline:
+
+1. **TOC Detection** — scans the first pages to locate the table of contents
+2. **TOC Extraction** — multi-pass extraction with completion verification
+3. **TOC → JSON** — structured flat list with hierarchy identifiers (`1`, `1.1`, `1.2.3`)
+4. **Physical Index Assignment** — verifies each section starts on the correct page
+5. **Tree Build** — nested tree structure with exact `start_index` + `end_index` per section
+6. **Summary Generation** — 1–2 sentence summary per node
+7. **Page Verification** — fuzzy-matches node titles against actual page text
+8. **Save** — stores the tree + metadata in Supabase
+
+### Chat & Retrieval (Flow 2)
+
+At query time, the LLM navigates the tree like a table of contents:
+1. Receives the full tree structure with section titles and summaries
+2. Selects the most relevant leaf nodes based on the query
+3. Fetches verbatim page content using exact `start_index → end_index` ranges
+4. Generates an answer grounded in the retrieved content
+
+The frontend receives the answer, the retrieved nodes with page ranges, and the LLM's tree-navigation reasoning — all displayed in the UI.
+
+---
+
+## Stack
+
+| Layer | Technology |
+|---|---|
+| Orchestration & Backend | **Lamatic AI** (4 flows — no external server) |
+| Storage | **Supabase** (PostgreSQL) |
+| Frontend | **Next.js 15** (App Router, Server Actions) |
+| Styling | **CSS custom properties** (dark-mode design system) |
+| SDK | **`lamatic`** npm package |
+
+---
+
+## Features
+
+- **PDF Upload** — drag-and-drop or paste a URL
+- **Tree-Structured RAG** — vectorless retrieval using hierarchical document index
+- **Multi-Turn Chat** — conversational history maintained across messages
+- **Chat Persistence** — conversations saved to `localStorage`, survive page navigations
+- **Interactive Tree Viewer** — explore the full document structure, nodes highlight on retrieval
+- **Source Panel** — view retrieved sections with page ranges and LLM reasoning
+- **Document Management** — list all documents, view trees, delete documents
+- **Markdown Rendering** — AI responses rendered with headings, lists, bold, code
+- **Responsive Dark UI** — premium design system with animations and micro-interactions
+
+---
+
+## Prerequisites
+
+- [Lamatic AI](https://lamatic.ai) account (free)
+- [Supabase](https://supabase.com) account (free tier)
+- Node.js 18+
+
+> **That's it.** No Groq account, no Railway, no Python environment needed.
+
+---
+
+## Setup
+
+### 1. Set Up Supabase
+
+Run this SQL in Supabase SQL Editor:
+
+```sql
+create table documents (
+  id uuid default gen_random_uuid() primary key,
+  doc_id text unique not null,
+  file_name text,
+  file_url text,
+  tree jsonb,
+  raw_text text,
+  tree_node_count integer default 0,
+  status text default 'completed',
+  created_at timestamptz default now()
+);
+alter table documents enable row level security;
+-- Only the Supabase service role (used server-side in Lamatic flows) can
+-- read and write documents. No direct client-side access is permitted.
+create policy "service_role_only" on documents
+  for all
+  using (auth.role() = 'service_role')
+  with check (auth.role() = 'service_role');
+```
+
+### 2. Import Lamatic Flows
+
+Import all 4 flows from the `flows/` folder into Lamatic Studio:
+
+| Flow | Folder | Purpose |
+|---|---|---|
+| Upload | `flows/flow-1-upload-pdf-build-tree-save/` | PDF → 7-stage pipeline → tree index → Supabase |
+| Chat | `flows/chat-with-pdf/` | Tree search → page fetch → LLM answer |
+| List | `flows/flow-list-all-documents/` | List all documents from Supabase |
+| Tree | `flows/flow-4-get-tree-structure/` | Return full tree JSON or delete a document |
+
+Add these secrets in **Lamatic → Settings → Secrets**:
+
+| Secret | Value |
+|---|---|
+| `SUPABASE_URL` | `https://xxx.supabase.co` |
+| `SUPABASE_ANON_KEY` | From Supabase Settings → API |
+| `SUPABASE_SERVICE_ROLE_KEY` | From Supabase Settings → API — **server-side only, never expose client-side** |
+
+> **Important:** `SUPABASE_SERVICE_ROLE_KEY` bypasses RLS. Store it in Lamatic Secrets only — never in `.env.local` shipped to the browser.
+
+### 3. Install and Configure
+
+```bash
+cd kits/assistant/pageindex-notebooklm
+npm install
+cp .env.example .env.local
+```
+
+Fill in `.env.local`:
+
+```env
+LAMATIC_API_KEY=...          # Lamatic → Settings → API Keys
+LAMATIC_PROJECT_ID=...       # Lamatic → Settings → Project ID
+LAMATIC_API_URL=...          # Lamatic → Settings → API Docs → Endpoint
+
+FLOW_ID_UPLOAD=...           # Flow 1 → three-dot menu → Copy ID
+FLOW_ID_CHAT=...             # Flow 2 → three-dot menu → Copy ID
+FLOW_ID_LIST=...             # Flow 3 → three-dot menu → Copy ID
+FLOW_ID_TREE=...             # Flow 4 → three-dot menu → Copy ID
+```
+
+### 4. Run Locally
+
+```bash
+npm run dev
+# → http://localhost:3000
+```
+
+---
+
+## Project Structure
+
+```text
+pageindex-notebooklm/  (TypeScript · Next.js/React)
+├── actions/
+│   └── orchestrate.ts        # TypeScript — Server actions — all 4 flow calls via Lamatic SDK
+├── app/
+│   ├── globals.css            # CSS — Design system (custom properties, animations)
+│   ├── layout.tsx             # TSX/React — Root layout with metadata
+│   └── page.tsx               # TSX/React — Main page — document list + chat + tree viewer
+├── components/
+│   ├── ChatWindow.tsx         # TSX/React — Chat UI with markdown, sources, persistence
+│   ├── DocumentList.tsx       # TSX/React — Document sidebar with search + delete
+│   ├── DocumentUpload.tsx     # TSX/React — Drag-and-drop / URL upload
+│   └── TreeViewer.tsx         # TSX/React — Interactive hierarchical tree viewer
+├── flows/
+│   ├── flow-1-upload-pdf-build-tree-save/
+│   ├── chat-with-pdf/
+│   ├── flow-list-all-documents/
+│   └── flow-4-get-tree-structure/
+├── lib/
+│   ├── lamatic-client.ts      # TypeScript — Lamatic SDK initialization
+│   └── types.ts               # TypeScript — Shared interfaces and types
+├── config.json                # Kit metadata
+└── .env.example               # Environment variable template
+```
+
+---
+
+## Deploying to Vercel
+
+```bash
+git checkout -b feat/pageindex-notebooklm
+git add kits/assistant/pageindex-notebooklm/
+git commit -m "feat: PageIndex NotebookLM — end-to-end Lamatic-powered tree RAG"
+git push origin feat/pageindex-notebooklm
+```
+
+Then in Vercel:
+1. Import your repo
+2. Set **Root Directory** → `kits/assistant/pageindex-notebooklm`
+3. Add all 7 env vars from `.env.local`
+4. Deploy
+
+---
+
+## Author
+
+**Saurabh Tiwari** — [st108113@gmail.com](mailto:st108113@gmail.com)
+GitHub: [@Skt329](https://github.com/Skt329)