design: full landing page and docs redesign — Precision Instrument aesthetic

- New design system: Syne display + IBM Plex Sans/Mono, electric amber accent
  oklch(0.78 0.17 78), near-black cool bg oklch(0.08 0.008 250)
- Replaced purple gradient SaaS aesthetic with data-forward, terminal-native look
- Hero: asymmetric split layout, benchmark numbers as instrument readings panel
- Features: spec-sheet layout (metric + description rows) replacing card grid
- Benchmark: animated before/after bar charts, real compression comparisons
- Providers: compact chip grid replacing 13 identical cards
- Quick Start: clean terminal blocks, no macOS chrome dots
- Removed all gradient text, glassmorphism, and section eyebrow on every section
- docs-style.css: full rebrand to match new tokens (amber accent, IBM Plex)
- Added PRODUCT.md (register: brand, Vercel-reference, anti-refs defined)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: vishal veerareddy

.impeccable/live/config.json

@@ -0,0 +1,8 @@
+{
+  "files": ["docs/**/*.html"],
+  "insertBefore": "</body>",
+  "commentSyntax": "html",
+  "cspChecked": true,
+  "framework": "multi-page-static",
+  "devServer": null
+}

PRODUCT.md

@@ -0,0 +1,37 @@
+# Product
+
+## Register
+
+brand
+
+## Users
+
+Solo developers and engineers evaluating AI coding gateway options. They arrive from GitHub, npm, or a referral link. They're already using Claude Code, Cursor, or Codex and are looking to cut costs or route through their own infrastructure. Primary action: `npm install -g lynkr`. They make the decision themselves; no lengthy sales process.
+
+## Product Purpose
+
+Lynkr is a self-hosted LLM gateway that compresses tokens before they reach the model. It sits between AI coding tools (Claude Code, Cursor, Codex) and 13+ LLM providers, reducing costs through smart tool selection (47-60%), TOON JSON compression (up to 87.6%), semantic caching (171ms hits, 0 tokens billed), and automatic complexity-based tier routing. Zero code changes required from the client tool.
+
+Success: a developer reads the landing page, understands the value in under 30 seconds, and runs the install command.
+
+## Brand Personality
+
+Precise, technical, trustworthy. Like Vercel: sharp dark theme, data-forward (real benchmark numbers, not marketing adjectives), confident copy that doesn't waste words. Every claim is specific and measurable.
+
+## Anti-references
+
+- Generic SaaS: Intercom/HubSpot-style soft gradients, stock illustrations, feature grids with "empower your workflow" copy.
+- Typical 2024 AI startup landing page: aurora gradient backgrounds, glowing cards, buzzword-dense hero copy.
+- shadcn/zinc-dark clones: deep zinc + purple but with nothing distinctive — looks like a template, not a product.
+
+## Design Principles
+
+1. **Show numbers, not adjectives.** Every capability claim is backed by a specific measured number. "87.6% fewer tokens" not "massive compression".
+2. **Earn trust through specificity.** Developers are skeptical. The copy reads like it was written by someone who actually ran the benchmark.
+3. **Speed of comprehension.** A developer should understand what Lynkr does, why it matters, and how to install it within 30 seconds of arrival.
+4. **Tool-native aesthetic.** The visual language should feel at home in a terminal-centric developer workflow. Sharp edges, monospace where appropriate, no decorative softness.
+5. **No fluff, no apology.** Don't qualify claims. Don't use filler intros. State the thing, then prove it.
+
+## Accessibility & Inclusion
+
+Best effort practical compliance. Body text must have sufficient contrast against dark backgrounds. No motion for users who prefer reduced motion.