A text-mode web browser with AI help, inspired by Lynx, plus TextFirst.css for text-first websites.
DBBasic TextBrowser: Ideas & Insights
Core Concept
A text-mode web browser with AI assistance that demonstrates how keyboard-first, intentional navigation can be superior to mouse-driven browsing.
Key Innovations
1. AI Integration in a Text Browser
Game Changer: While Lynx is lightweight and fast, it has no intelligence. Adding GPT-5 gives you:
- Natural language navigation: "go to wikipedia for WebDAV"
- Content extraction: "find the pricing information"
- Summarization: "what are the main points?"
- Translation: "translate this to Spanish"
- Q&A: "what does this article say about security?"
- Function calling: AI can actually navigate the browser, not just suggest URLs
Why it works in text mode:
- Clean, structured content (no CSS/JS soup)
- AI gets exactly what the user sees
- Natural fit with command-driven interface
- Faster than most modern websites render
2. Numbered Links
The Insight: Every clickable element should be numbered and accessible via keyboard.
Current Implementation:
- Links numbered inline:
[0] Main page,[1] Talk, etc. - Press 0-9 for instant access to first 10 links
- Press G + number for any link
- Link list at bottom of page
Why this matters: No mouse hunting, no accidental clicks, deliberate navigation.
3. Form Support
Features:
- Automatic form detection
- Press F to fill out forms
- Interactive field-by-field input
- Supports GET and POST methods
- Works with search engines (DuckDuckGo, etc.)
4. Text-First Design Philosophy
Core Principle: Build for text/keyboard first, add GUI as progressive enhancement.
Advantages:
- Forces you to think about information hierarchy
- Accessibility built in from day one
- No accidental interactions
- Fast and intentional
- Works over SSH, slow connections, screen readers
Modern web dev typically goes:
- Design visual layout
- Add interactions
- Try to make it accessible
- Try to make it fast
We went:
- What's the actual content/function?
- How do I navigate with keyboard?
- How do I make it smart with AI?
The Big Idea: Numbered Navigation in Modern Browsers
The Problem
Firefox, Chrome, Safari all assume mouse-first navigation. Keyboard users are second-class citizens.
The Solution
What if Firefox had native numbered navigation?
- Press a key (like
/orF) - Every clickable element gets a number overlay
- Type the number, hit Enter
- Element is clicked
Benefits
- Accessibility: Motor disabilities, screen readers
- Power users: Way faster than mouse
- Remote desktop: Mouse over RDP is painful
- Laptop trackpads: Trackpads are terrible
- Gaming/streaming: Keep hands on keyboard
- Efficiency: No visual hunting
Existing Attempts (All Extensions)
- Vimium / Vimium-C: Press
f, links get letter combos - Tridactyl: Vim-like navigation
- SurfingKeys: Chrome equivalent
Problem: They're extensions, not core browser features.
Why Browsers Don't Do This
- Mouse-first mindset
- Fear of complexity ("normal users won't use it")
- Design language assumes hover states
- "Someone will make an extension"
But: This should be core browser functionality, just like Ctrl+T, Ctrl+L, Ctrl+F.
Future Features to Add
High Priority
- [ ] History/Back button: Navigate backward through visited pages
- [ ] Bookmarks: Save and manage favorite pages
- [ ] Search within page:
/to search like vim - [ ] Better table rendering: Parse HTML tables into ASCII tables
AI Enhancements
- [ ] Web search function: Give AI ability to search the web
- [ ] Extract to file: "save this table as CSV"
- [ ] Page comparison: "compare these two product pages"
- [ ] Multi-page chains: AI visits multiple pages to gather info
- [ ] Content filtering: "show only the article, hide everything else"
Navigation Improvements
- [ ] Tab support: Multiple pages open at once
- [ ] Session restore: Reopen tabs from last session
- [ ] Link previews: Hover (or key) to see URL before visiting
- [ ] Smart back: Remember scroll position
Downloads
- [ ] File download detection: Detect and save binary files
- [ ] Download manager: Track download progress
- [ ] Auto-open: Open downloaded files in appropriate apps
Performance
- [ ] Caching: Cache pages locally
- [ ] Prefetch: Preload likely next pages
- [ ] Compression: Support gzip/brotli
Technical Insights
What We Learned
- Text wrapping is critical: Setting
body_width=0broke everything - GPT-5 API changes: New models need
max_completion_tokensinstead ofmax_tokens, and don't support customtemperature - Form handling is complex: Need to handle relative URLs, GET vs POST, field types
- Link extraction requires filtering: Skip
#,javascript:,mailto:links - Color makes a huge difference: Even in a text browser, syntax highlighting helps
Architecture Decisions
- BeautifulSoup for HTML parsing (could switch to lxml for speed)
- html2text for HTML→markdown conversion (works well but could be customized)
- curses for terminal UI (standard but limiting on Windows)
- requests for HTTP (simple but no async support)
Performance Characteristics
- Faster than graphical browsers for simple pages
- AI calls are the bottleneck: GPT-5-nano is slower than expected
- No JS execution: Some sites won't work (SPAs, etc.)
- Text rendering is instant: No layout engine, no reflows
Potential Projects
1. Firefox Extension: Native Numbered Links
Build a better Vimium that:
- Uses actual numbers (not letter combos)
- Integrates with browser UI
- Has better visual design
- Supports forms, buttons, all clickables
2. Text-First Web Framework
A web framework that generates both:
- Text-mode interface (for this browser)
- GUI interface (progressive enhancement)
Like how GraphQL has one schema, multiple clients.
3. AI-Enhanced Bookmarklet
A bookmarklet that adds AI to any browser:
- Summarize page
- Extract data
- Answer questions
- Navigate intelligently
4. Terminal Browser as Developer Tool
Use this as a debugging tool:
- See what screen readers see
- Test keyboard navigation
- Validate content hierarchy
- Check accessibility
Philosophy
Text-Only Forces Good Design
When you can't hide behind visual design:
- Information hierarchy must be clear
- Navigation must be intuitive
- Content must be well-structured
- Every interaction must be intentional
Keyboard-First is Faster
Once you learn the keys:
- No visual hunting for clickable areas
- No mouse precision required
- No accidental clicks
- Muscle memory develops
AI Changes Everything
AI in a browser isn't just a chatbot:
- It can navigate for you
- It can extract information
- It can answer questions about what you're viewing
- It can chain actions together
The browser becomes an agent, not just a viewer.
Success Metrics
How do we know this approach works?
- Speed: From thought to action is faster than mouse
- Intentionality: Every action is deliberate
- Accessibility: Works for everyone, not just mouse users
- Intelligence: AI makes browsing smarter, not just automated
- Simplicity: Core functionality is obvious and learnable
Related Work
- Lynx: The original text browser (no AI)
- w3m: Another text browser (better table support)
- elinks: Feature-rich text browser
- Vimium: Vim keybindings for browsers
- qutebrowser: Keyboard-focused browser
- surf: Minimalist browser
- Browserless: Headless browser automation
Questions to Explore
- Could this approach work for mobile?
- What about voice navigation with numbers?
- How would numbered navigation work with dynamic content?
- Could AI predict which link you want?
- What if every app had a text-mode interface?
Contact & Contributions
This is an exploration of what browsers could be. Ideas, code, and discussion welcome.
Core insight: Text-first, keyboard-first, AI-enhanced browsing might actually be the future, not the past.