Fauxmium is a local Apache-2.0 licensed HTTP/HTTPS proxy and autonomous browser runtime. It intercepts web requests and dynamically synthesizes full-fidelity HTML, procedural 16-bit PCM WAV audio, vector graphics, and Veo video streams on the fly with multi-model AI and site-level design coherence.
Experience how Fauxmium generates continuous websites, persists domain design DNA, morphs aesthetics via Browsing Lenses, and synthesizes procedural audio in real time.
Fauxmium pairs prompt prefix caching, procedural audio generation, and site memory to make AI web exploration fast, cohesive, and economical.
Synthesizes HTML, vector SVG, procedural WAV audio, and video streams in a unified pipeline. The server dynamically routes requests to appropriate AI models or fast local synthesizers.
Transform any website into alternate temporal or aesthetic worldviews (Retro 90s, Cyberpunk 2077, Neo-Brutalist, Film Noir, Arcade 8-Bit) across typography, layout, color palette, and audio.
Solves LLM amnesia by capturing navbar structure, brand tokens, and CSS variables across pages. When clicking a link, the originating card snippet is injected to guarantee logical continuity.
Structures invariant system prompts (>1,024 tokens) and tiers viewport dimensions to 64px buckets. Delivers up to 75% LLM cost reduction and sub-second generation latencies.
Implements standard HTTP 206 Partial Content range requests for video playback. Supports timeline seeking, chunk streaming, and companion poster frame visual seeding.
Redesigned Chrome Extension displays live session token counts, prompt caching discounts, multi-modal asset breakdowns, and hot-swaps Browsing Lenses without restarting the server.
Fauxmium sits between Chrome and the internet, intercepting every HTTP/HTTPS request and translating it into structured, cached prompts.
[Browser / Chrome]
│
▼ (HTTP / HTTPS Request)
[Fauxmium Proxy Server (127.0.0.1:3991)]
│
├── Accept: text/html
│ ├── Invariant System Prompt (>1,024 tokens) ───► [PROMPT CACHE HIT: 75% OFF]
│ ├── Decomposed Route + Site Memory (CSS/Brand)
│ ├── Link Intent Context (Clicked Card + Target URL)
│ └── AI LLM (Gemini 3.8 / GPT-6 / Claude 5.5) ──► Complete HTML Page
│
├── Accept: image/* or /image/
│ ├── Dimensions, Role (hero, icon, thumbnail), Brand Colors
│ └── Imagen 4 / DALL-E 3 / SVG Placeholder ─────► Vector/Raster Image
│
├── Accept: video/* or /video/
│ ├── Camera Motion, Aspect Ratio, Seed Poster Frame
│ ├── Veo 3.1 Fast / Local Fallback MP4
│ └── HTTP 206 Byte-Range Response ──────────────► Seeking Video Stream
│
└── Accept: audio/* or /audio/
├── Audio Type (chiptune, synth, speech, ambient), BPM, Mood
└── Zero-Dependency Node.js PCM WAV Synth ─────► 16-Bit Audio Buffer
Both Anthropic and Gemini cache prompt prefixes when the invariant prefix exceeds 1,024 tokens. Fauxmium groups the system prompt into a static block followed by 64px viewport tiers (e.g. 1920×1080), avoiding cache busting when browser windows resize slightly.
When --audio-provider synth is active, Fauxmium generates 16-bit uncompressed PCM WAV files directly in Node.js using Math.sin, ADSR envelopes, low-pass filters, and formant speech synthesis. No external binaries or paid APIs needed.
Configure your multi-modal runtime parameters and generate ready-to-run commands with appropriate environment variables.
npx fauxmiumGEMINI_API_KEY=your_gemini_api_keyFauxmium's prompt engine instructs models to author declarative media URLs with query parameters that guide image framing, video cinematography, and audio characteristics.
| Tag / MediaType | Supported Query Parameters | Description & Behavior | Example Usage |
|---|---|---|---|
<img> |
description, width, height, role, aspect_ratio |
Guides image composition. role (hero, logo, icon, avatar, thumbnail) influences shot framing and detail density. Inherits site CSS brand palette.
|
/img/hero.jpg?description=retro+cyberdeck&width=1280&height=720&role=hero |
<video> |
description, width, height, camera, poster_description, aspect_ratio |
Configures Veo 3.1 video generation. camera specifies motion (dolly, pan, static, orbit). poster_description seeds the companion poster frame for visual continuity.
|
/video/demo.mp4?description=quantum+engine+spin&camera=orbit+slow&poster_description=core+offline |
<audio> |
description, type, bpm, mood, transcript, voice |
Generates procedural WAV audio or TTS. type (chiptune, synth, speech, ambient, sound-effect) dictates synth oscillator shapes. transcript scales speech duration.
|
/audio/bg.wav?type=chiptune&bpm=140&mood=energetic&description=boss+fight+stage |
Configured in config/models.json, supporting dynamic prefix caching discounts and multi-modal per-unit accounting.
| Provider | Model Name | Role | Input / Cached / Output (per 1M) | Media Per Unit |
|---|---|---|---|---|
| Gemini | gemini-3.8-flash (Default) |
Ultra-fast HTML synthesis | $0.15 / $0.0375 / $0.60 | — |
| Gemini | gemini-3.8-pro |
Complex logic & dense layout | $1.25 / $0.3125 / $5.00 | — |
| Gemini | gemini-3.1-flash-image (Default) |
Fast multi-modal image synthesis | — | $0.039 / img |
| Gemini | veo-3.1-fast-generate-preview |
High-definition video clips | — | $0.400 / sec |
| OpenAI | gpt-6-sol (Default) |
Balanced reasoning & code | $2.00 / $0.50 / $8.00 | — |
| OpenAI | gpt-6-luna |
Low-latency micro-tier | $0.30 / $0.075 / $1.20 | — |
| OpenAI | gpt-image-2.5 |
High-fidelity raster images | — | $0.040 / img |
| Anthropic | claude-sonnet-5-5 (Default) |
Flawless semantic HTML | $3.00 / $0.75 / $15.00 | — |
| Anthropic | claude-haiku-5-5 |
High-throughput streaming | $0.80 / $0.20 / $4.00 | — |
| Groq | openai/gpt-oss-120b (Default) |
Open-weight LPU acceleration | $0.15 / $0.0375 / $0.60 | — |
| Local Synth | synth-wav |
Procedural 16-bit PCM Audio | Free (Local CPU) | $0.000 / wav |