Azul – AI-Powered Terminal Web Browser

Background

Azul started as an experiment: what if web browsing could happen entirely in the terminal, controlled by natural language? The result is a full TUI (Terminal User Interface) web browser that combines traditional keyboard navigation with an AI assistant capable of executing browsing commands.

Core Features

Terminal-First Browsing

Azul renders web content in the terminal using Ratatui, with:

• Full keyboard navigation for links, scrolling, and tab management
• JavaScript rendering via headless Chrome when needed
• Bookmarks and history persistence with SQLite
• Dark cosmic theme optimized for terminal aesthetics

AI Chat Integration

The AI panel supports natural language commands that translate to browser actions:

• Navigation: "go to wikipedia.org" or "search for Rust programming"
• Interaction: "click the first link" or "scroll down"
• Research: "summarize this page" or "find information about X"

The AI uses tool-calling to execute these actions within the browser context, making it possible to browse hands-free.

Multi-Engine Search

Built-in search across six engines:

1. DuckDuckGo – Privacy-focused general search
2. Wikipedia – Direct encyclopedia queries
3. arXiv – Academic paper search
4. PubMed – Medical/scientific literature
5. Google Scholar – Academic citations
6. OpenLibrary – Book search

Users can switch engines on the fly or let the AI choose based on query context.

Technical Architecture

┌─────────────────────────────────────┐
│           Terminal UI (Ratatui)     │
├─────────────┬───────────────────────┤
│  Browser    │      AI Chat          │
│  Panel      │      Panel            │
├─────────────┴───────────────────────┤
│         Core Engine (Rust)          │
├─────────────┬───────────────────────┤
│  Headless   │    AI Provider        │
│  Chrome     │    (OpenAI/Ollama)    │
├─────────────┴───────────────────────┤
│         SQLite (Storage)            │
└─────────────────────────────────────┘

AI Provider Flexibility

Azul supports multiple AI backends:

• OpenAI – GPT-4 and GPT-3.5 via API
• OpenRouter – Access to multiple models
• Ollama – Local LLM execution for privacy
• Custom endpoints – Any OpenAI-compatible API

Configuration is handled via JSON, allowing users to switch providers without code changes.

Implementation Details

Tool-Calling System

The AI assistant uses a structured tool-calling interface:

pub enum BrowserTool {
    Navigate { url: String },
    Search { query: String, engine: SearchEngine },
    Click { selector: String },
    Scroll { direction: ScrollDirection, amount: u32 },
    ExtractText { selector: Option<String> },
    Screenshot,
}

When the AI decides to execute an action, it returns a tool call that the browser interprets and executes, then reports the result back for continued conversation.

Keyboard Shortcuts

Power users can bypass AI entirely with extensive keybindings:

• g – Go to URL
• / – Search current page
• j/k – Scroll up/down
• Tab – Cycle through links
• Enter – Follow link
• b – Open bookmarks
• H – View history
• t – New tab

Optional RAG Integration

For research workflows, Azul supports RAG (Retrieval-Augmented Generation):

1. Index pages you visit into a local vector store
2. Query your browsing history semantically
3. Get AI responses grounded in pages you've read

Challenges Solved

Terminal Rendering

Rendering web pages in a terminal is inherently lossy. Azul handles this by:

• Stripping complex CSS while preserving semantic structure
• Converting images to ASCII art or placeholder text
• Maintaining link relationships for navigation

AI Latency

To keep browsing responsive:

• Tool calls execute immediately without waiting for full AI response
• Streaming responses show progress in real-time
• Local LLM option (Ollama) eliminates network latency

Cross-Platform Support

Rust's cross-compilation made it straightforward to ship binaries for:

• Linux x86_64
• macOS x86_64
• macOS ARM64 (Apple Silicon)

Results

Azul demonstrates that terminal-based, AI-augmented browsing is not just possible but practical for certain workflows:

• Research: Navigate academic papers hands-free while taking notes
• Accessibility: Voice-to-text combined with AI commands enables hands-free browsing
• Privacy: Local LLM option keeps all data on-device
• Efficiency: Power users can browse faster than traditional GUI browsers

Future Directions

• Session sync across devices
• Plugin system for custom tools
• Voice input integration
• Collaborative browsing with shared sessions

Azul is open source and available at github.com/0xSero/Azul.