HTML Chat

A lightweight, zero-runtime-dependency web chat client for OpenAI-compatible APIs that builds into a single self-contained index.html file. Includes local IndexedDB storage, client-side RAG with vector search, customizable JavaScript pipeline hooks, and local code execution.

---

Features

• Single-File Build: Compiles entire app (JS, CSS, Prism highlighting, KaTeX math fonts) into one portable index.html.
• API Compatible: Works with OpenAI, Ollama, OpenRouter, LocalAI, vLLM, LM Studio, and any OpenAI-compatible endpoint.
• Client-Side RAG:
  • File upload with background vector embedding and batch processing.
  • Cosine-similarity retrieval with customizable token limits and similarity thresholds.
  • Custom JavaScript hooks for chunking, context retrieval, deduplication, and chunk merging.
  • Import/export of chunk and vector datasets.
• God Mode (JS Execution): Agentic feedback loop allowing models to output <run> blocks that execute directly in the browser and feed results back to the model.
• Rich Rendering: Markdown, KaTeX math typesetting ($inline$, $$display$$), and Prism.js syntax highlighting.
• IndexedDB Storage: Conversations, messages, files, vector embeddings, and preferences persist locally in your browser.

---

Quick Start

Running the pre-built file

Open index.html in any modern web browser. No server required.

Development

# Install dependencies
npm install

# Start development server
npm run dev

# Build single-file production bundle to ./index.html
npm run build

# Preview build
npm run preview

---

Configuration

1. Click Settings in the sidebar.
2. Provide your Base URL (default: https://api.openai.com/v1) and API Key.
3. Specify available model names as a comma-separated list (e.g., gpt-4o, gpt-4o-mini, llama3).
4. (Optional) Set an Embeddings Model (e.g., text-embedding-3-small) in Super Secret Settings to enable vector search features.

---

Shortcuts & Actions

Keyboard Shortcuts

Shortcut	Action
Ctrl + Enter / Cmd + Enter	Send message / Save edits
Alt + T	New chat
Alt + W	Delete current chat
Alt + R	Rename current chat
Alt + P	Toggle sidebar visibility
Alt + O	Toggle header visibility
Alt + I	Open Super Secret Settings
Alt + Up / Alt + Down	Switch active chat up / down
Shift + Up / Shift + Down	Scroll between message blocks
Shift + Enter (on Send)	Append user message without making an API call

Modifier Clicks

• Ctrl / Cmd + Click on Code / Math: Copy raw content directly to clipboard.
• Ctrl / Cmd + Click on Chat Title: Copy full chat transcript as Markdown.
• Ctrl / Cmd + Click on File Item: Open Advanced RAG Settings for that file.
• Ctrl / Cmd + Click on Settings Header: Open Super Secret Settings.
• Alt + Click on File Item: Replace file contents via file picker.
• Alt + Click on Chat Item: Export specific chat as JSON.
• Alt + Click on "+ New": Import chats from JSON.

---

Architecture

The codebase contains no external runtime frameworks and enforces unidirectional dependencies:

src/scripts/
├── core/       # Pure functions: formatting, tokens, pipeline, progress, vector math
├── data/       # IndexedDB repositories and storage keys
├── store/      # In-memory application state and event emitter
├── services/   # API clients, conversation loops, RAG, and execution engines
├── ui/         # DOM manipulation, components, views, and markdown rendering
└── app/        # Command registry, event delegation, shortcuts, and bootstrap