Fauxmium — The Infinite Generated Web

Everything you see inside the Fauxmium browser is generated on the fly. It is not real.

This project is a proof‑of‑concept showing how generative AI can create an effectively infinite “web” as you browse. Fauxmium launches Chrome, intercepts navigations and image requests, and routes them to a local proxy that asks AI models to generate HTML and images for the requested URL.

Not affiliated with any employer or product.

Quick Start

Run with default settings (Google/Gemini for text and images):

npx fauxmium

Recommended: open DevTools

npx fauxmium --devtools

Requirements

• Node.js
• API keys for the providers you intend to use (see Environment Variables below)

Environment Variables

Place in a .env file at project root or export via your shell:

• GEMINI_API_KEY or GOOGLE_API_KEY — for Google text and images
• OPENAI_API_KEY — for OpenAI text
• ANTHROPIC_API_KEY — for Anthropic text
• GROQ_API_KEY — for Groq text

The CLI automatically resolves keys from the environment. You can override with flags (--api-key, --image-api-key, --video-api-key).

Example .env:

GEMINI_API_KEY=your_gemini_key
OPENAI_API_KEY=your_openai_key
ANTHROPIC_API_KEY=your_anthropic_key
GROQ_API_KEY=your_groq_key

CLI Usage

Default command (no subcommand) uses Google/Gemini text and Google/Gemini images.

npx fauxmium [options]

Provider commands for text:

npx fauxmium gemini   [options]
npx fauxmium google   [options]   # alias of gemini
npx fauxmium openai   [options]
npx fauxmium anthropic [options]
npx fauxmium groq     [options]

Common options:

• --hostname, -H Default: 127.0.0.1
• --port, -p Default: 3001
• --devtools Open DevTools on launch (default: false)
• --model, -m Text model (provider-specific defaults/choices)
• --api-key Explicit API key for text provider (overrides env)
• --image-provider Image provider (currently gemini/google only)
• --image-model, -i Image model (e.g., gemini-2.5-flash-image-preview)
• --image-api-key Explicit API key for image provider (overrides env)
• --video-provider Video provider (currently gemini/google only)
• --video-model, -v Video model (e.g., veo-3.0-fast-generate-preview)
• --video-api-key Explicit API key for video provider (overrides env)

Image configuration:

• Images are currently supported only via Google/Gemini.
• You can configure image settings via:
  • the same provider command's nested images subcommand, or
  • the top-level images command with provider subcommands, or
  • the options on the main command.

Video configuration:

• Videos are currently supported only via Google/Gemini.
• You can configure video settings via:
  • the same provider command's nested videos subcommand, or
  • the top-level video command with provider subcommands, or
  • the options on the main command.

Examples:

# Default: Gemini text + images + videos
npx fauxmium

# OpenAI for text, Gemini for images and videos
npx fauxmium openai --api-key $OPENAI_API_KEY --image-api-key $GEMINI_API_KEY --video-api-key $GEMINI_API_KEY

# Change port/host and open DevTools
npx fauxmium -p 8080 -H 127.0.0.1 --devtools

# Choose text, image, and video models explicitly (Gemini)
npx fauxmium gemini -m gemini-2.5-flash -i gemini-2.5-flash-image-preview -v veo-3.0-fast-generate-preview

# Use top-level image command for image-specific configuration
npx fauxmium images gemini --image-model gemini-2.5-flash-image-preview

# Use top-level video command for video-specific configuration
npx fauxmium video gemini --video-model veo-3.0-generate-001

# Use nested subcommands for configuration
npx fauxmium anthropic images --image-provider gemini --image-model gemini-2.5-flash-image-preview
npx fauxmium anthropic videos --video-provider gemini --video-model veo-3.0-fast-generate-preview

For full help and the list of default models per provider:

npx fauxmium --help

How It Works

High-level flow:

1. Fauxmium launches headful Chrome via Puppeteer.
2. It installs a bundled MV3 extension at runtime and configures it with the proxy host/port.
3. All page navigations and image requests are intercepted.
4. Intercepted requests are redirected to a local proxy server that asks AI models to generate HTML or images based on the requested URL.
5. The browser renders the returned content.

Components:

• Browser Controller (browser.js)
  • Headful Chrome launch, runtime extension install, configuration via CDP Extensions.setStorageItems.
  • Intercepts requests per page.
  • Shows a warning page on startup.
• Proxy Server (server/index.js)
  • Endpoints:
    • GET /html — Streams AI‑generated HTML (extracted from ```html fences).
    • GET /image — Returns AI‑generated image bytes (PNG/JPEG/etc.) or a 1×1 transparent PNG on error.
    • GET /cost — In‑memory session usage/cost summary (used by the extension).
  • Loads model pricing from Helicone for cost tracking (loadCosts).
• Text/Image Generation (/lib)
  • lib/aiAdapter.js — Text streaming via Vercel AI SDK across providers (Google/OpenAI/Anthropic/Groq). Images via @google/genai (Google only).
  • lib/costCalculator.js — Tracks token usage and cost per request/session.
  • lib/processChunks.js — Applies processors and manages stream flush with an END sentinel.
  • lib/streamCodeBlocks.js — Extracts ```html fenced content from the text stream.
  • lib/prompts.js — Loads and interpolates prompts/html.txt and prompts/image.txt.
• Chrome Extension (extension/)
  • Popup fetches /cost and displays total and per‑request costs.

Prompts:

• prompts/html.txt
• prompts/image.txt

These are read from disk on each request, so you can tweak prompts without restarting.

Constraints and Behavior

• Subresources:
  • Only navigation (page) and image requests are forwarded to the proxy.
  • Other subresources (e.g., external CSS/JS) are blocked.
  • Generate inline CSS and JS within the HTML.
• State:
  • Navigations are stateless; there is no cross‑page memory at present.
• Images and Videos:
  • Only Google/Gemini image and video generation is supported at this time.
• Costs and usage:
  • Pricing is fetched from Helicone per model; unknown models default to 0.
  • Some providers may not report usage; such requests are counted as 0 cost.
  • Image and video generation costs are not currently tracked.
• Headers:
  • Referer is stripped (set to empty) for proxy‑bound requests as a temporary workaround.

Roadmap (abridged)

• Improve image prompts with width/height/description parsed from image URLs.
• Optional session memory to carry context across navigations.
• Add additional image providers behind a common interface.
• Add /reset-costs endpoint and refresh/reset in the extension popup.
• Better error handling, timeouts, structured logs, and tests.

License

Apache‑2.0 — see LICENCE.

Contributing

Issues and PRs are welcome. Please describe changes clearly and include repro steps where relevant.