Llamaindex — AI Agents for Document OCR + Workflows

# Build brief — an original marketing website in the taste of Llamaindex

> Reference: https://www.llamaindex.ai · UIDiscovery: https://uidiscovery.com/sites/llamaindex · captured 2026-09-23 at a 1440px desktop viewport (+ 390px iPhone shots) · [full page](https://uidiscovery.com/media/captures/llamaindex/full.png) · [scroll video](https://uidiscovery.com/media/captures/llamaindex/preview.mp4)

## 1. Goal
Build an original marketing website for **[the user's brand]** that feels as considered as Llamaindex — same structure, rhythm and polish, entirely new identity and content.

## 2. Originality rules
- Build an ORIGINAL design for the user's own brand in the same *taste* as Llamaindex: structure, rhythm, hierarchy, palette mood and type pairing.
- Never reuse Llamaindex's logo, name, product imagery, illustrations, icons, screenshots or copy. Write new copy for the user's product.
- If the user hasn't said what their product is, ask first (or use clearly marked placeholders like [Brand], [Product], [Audience]).
- Reference screenshots are for looking at, not for embedding or tracing.

## 3. The taste, in short
- Llamaindex (llamaindex.ai): Developer Tools marketing website, light, minimal.
- 16 sections: Logo cloud → Content → Stats → Statement → Content → Features → Statement → Content → Features → Content → Features → Content.
- Spacing: about a screen per idea (median section ≈541px).
- Type: one family (Overused Grotesk). Imagery: mostly typography with little or no imagery, some photography.

## 4. Page structure (top to bottom)

### 1. Logo cloud — borrow trust from recognisable customers or partners.
- Layout: a single row or wall of monochrome marks with a one-line lead-in. Tall, ≈1.8 screens (1657px at 1440), page background; 31 image elements (icons, logos, product UI — not 31 photos); 8 text links instead of buttons.
- Slots: One-line lead-in (e.g. “Trusted by…”); 6–10 monochrome customer/partner marks (your own, with permission).
- Topic (tone/length only — never reuse): “Turn Any Document Into AI-Ready Context”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-01.webp)

### 2. Content — supporting content that deepens one idea.
- Layout: text + media block. Compact band (285px at 1440), page background; no images (type-led); no interactive controls.
- Slots: Heading (~6 words, like the reference); Body copy.
- Topic (tone/length only — never reuse): “We parse your most complex docs”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-02.webp)

### 3. Content — supporting content that deepens one idea.
- Layout: text + media block. Half to one screen (541px at 1440), page background; 8 images/icons (a set of cards or a small gallery); 3 buttons (tabs, toggles or per-card actions).
- Slots: Heading; Body copy; Supporting visual (see image plan); Primary + secondary action.
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-03.webp) · imagery: slot 1

### 4. Stats — prove scale or impact with a few big numbers.
- Layout: 3–4 large numerals with short labels in a row. Compact band (191px at 1440), page background; no images (type-led); no interactive controls.
- Slots: 3–4 numbers with short labels; Source/footnote if claims need it.
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-04.webp)

### 5. Statement — a large-type moment that sets tone and pace.
- Layout: oversized typography with lots of space. Compact band (339px at 1440), page background; no images (type-led); no interactive controls.
- Slots: Heading (~5 words, like the reference); Body copy.
- Topic (tone/length only — never reuse): “Unrivaled performance across complex documents”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-05.webp)

### 6. Content — supporting content that deepens one idea.
- Layout: text + media block. Half to one screen (606px at 1440), own light band #fafafa; 6 images/icons (a set of cards or a small gallery); 3 buttons (tabs, toggles or per-card actions).
- Slots: Heading; Body copy; Supporting visual (see image plan); Primary + secondary action.
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-06.webp) · imagery: slot 1

### 7. Features — explain the key benefits, grouped so each can be scanned in seconds.
- Layout: section title over a grid or alternating rows of benefit items. Half to one screen (754px at 1440), page background; no images (type-led); 2 text links instead of buttons; 5 headings → ~4 items under a title.
- Slots: Section title (~3 words, like the reference); 4 items: icon or visual + short title + 1–2 line description; Text link action.
- Topic (tone/length only — never reuse): “Parse”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-07.webp)

### 8. Features — explain the key benefits, grouped so each can be scanned in seconds.
- Layout: section title over a grid or alternating rows of benefit items. Half to one screen (680px at 1440), page background; 4 images/icons (a set of cards or a small gallery); 2 text links instead of buttons; 4 headings → ~3 items under a title.
- Slots: Section title (~3 words, like the reference); 3 items: icon or visual + short title + 1–2 line description; Text link action.
- Topic (tone/length only — never reuse): “Fully open-source”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-08.webp)

### 9. Statement — a large-type moment that sets tone and pace.
- Layout: oversized typography with lots of space. Compact band (354px at 1440), page background; no images (type-led); no interactive controls.
- Slots: Heading (~6 words, like the reference); Body copy.
- Topic (tone/length only — never reuse): “How leading teams use document intelligence”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-09.webp)

### 10. Content — supporting content that deepens one idea.
- Layout: text + media block. Half to one screen (618px at 1440), page background; 6 images/icons (a set of cards or a small gallery); 3 text links instead of buttons.
- Slots: Heading; Body copy; Supporting visual (see image plan); Text link action.
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-10.webp) · imagery: slot 1

### 11. Features — explain the key benefits, grouped so each can be scanned in seconds.
- Layout: section title over a grid or alternating rows of benefit items. Tall, ≈1.4 screens (1300px at 1440), page background; no images (type-led); 2 text links instead of buttons.
- Slots: Section title (~5 words, like the reference); 3 items: icon or visual + short title + 1–2 line description; Text link action.
- Topic (tone/length only — never reuse): “Built for every document-heavy industry”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-11.webp)

### 12. Content — supporting content that deepens one idea.
- Layout: text + media block. Compact band (366px at 1440), own light band #f5f5f5; 2 image/graphics; 1 text link instead of buttons.
- Slots: Heading; Body copy; Supporting visual (see image plan); Text link action.
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-12.webp) · imagery: slot 1

### 13. Content — supporting content that deepens one idea.
- Layout: text + media block. Compact band (366px at 1440), own light band #f5f5f5; 2 image/graphics; 1 text link instead of buttons.
- Slots: Heading; Body copy; Supporting visual (see image plan); Text link action.
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-13.webp) · imagery: slot 1

### 14. Features — explain the key benefits, grouped so each can be scanned in seconds.
- Layout: section title over a grid or alternating rows of benefit items. About one full screen (935px at 1440), own light band #f5f5f5; 10 images/icons (a set of cards or a small gallery); 1 text link instead of buttons.
- Slots: Section title (~3 words, like the reference); 9 items: icon or visual + short title + 1–2 line description; Text link action.
- Topic (tone/length only — never reuse): “Enterprise Ready”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-14.webp)

### 15. Content — supporting content that deepens one idea.
- Layout: text + media block. Compact band (285px at 1440), own light band #f5f5f5; no images (type-led); no interactive controls.
- Slots: Heading (~5 words, like the reference); Body copy.
- Topic (tone/length only — never reuse): “Trusted by leading AI teams”
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-15.webp)

### 16. Content — supporting content that deepens one idea.
- Layout: text + media block. Compact band (372px at 1440), own light band #f5f5f5; 6 images/icons (a set of cards or a small gallery); no interactive controls.
- Slots: Heading; Body copy; Supporting visual (see image plan).
- [Screenshot](https://uidiscovery.com/media/captures/llamaindex/s-16.webp) · imagery: slot 2

Inner pages also captured (Pricing): https://uidiscovery.com/sites/llamaindex?view=pages

## 5. Design system
### Colour
| Role | Hex | Use | Contrast on background |
|---|---|---|---|
| background | `#ffffff` | page background | — |
| foreground | `#000000` | body text, headings | 21.0:1 |
| accent | `#000000` | none captured — monochrome. Primary buttons invert (foreground fill, background text); add your brand colour only for small highlights | — |
| surface | `#fafafa` | cards, inputs, alternate bands | text on it 20.1:1 |
| muted | `#7f7f7f` | secondary text, borders | 4.0:1 — borders / large text only |

Captured palette: `#ffffff` `#000000` `#f5f5f5` `#7f7f7f` `#5f5f5f` `#fafafa`. Section bands top→bottom: `#ffffff` → `#fafafa` → `#f5f5f5`. Roles are assigned automatically — sanity-check them against the screenshot. Keep the palette's *mood* (light, neutral, desaturated) and proportions; swap the accent for the user's brand colour.

### Tokens
```css
:root {
  --background: #ffffff;
  --foreground: #000000;
  --accent: #000000;
  --surface: #fafafa;
  --muted: #7f7f7f;
  --font-heading: "Inter", ui-sans-serif, system-ui, sans-serif;
  --font-body: "Inter", ui-sans-serif, system-ui, sans-serif;
  --radius: 12px; /* calibrate from the screenshot */
}

/* Tailwind v4 — in globals.css after @import "tailwindcss"; */
@theme {
  --color-background: var(--background);
  --color-foreground: var(--foreground);
  --color-accent: var(--accent);
  --color-surface: var(--surface);
  --color-muted: var(--muted);
  --font-heading: var(--font-heading);
  --font-sans: var(--font-body);
}
```

### Typography
- Headings: **Inter** (Google Fonts) — reference uses Overused Grotesk. “Overused Grotesk” is likely a custom or licensed face. Inter is a neutral stand-in — pick a closer free match by eye from the screenshot.
- Body: same family — a single-family system; create contrast with size and weight, not a second face.
- Scale (starting point, sizes weren't measured — calibrate against the 1440px screenshot): display clamp(2.5rem, 5vw, 4.25rem), h2 clamp(1.75rem, 3vw, 2.5rem), h3 1.25rem, body 1rem–1.125rem / 1.6, small 0.875rem. Headings: tight tracking (-0.02em), line-height 1.05–1.15.

### Spacing & rhythm
Sections run ≈541px tall at 1440 (median; range 191–1657px, captures cap at 1700px) — roughly a screen per idea. Start with 96–120px vertical padding per section on desktop and about two-thirds of that on mobile. The background changes 3 times down the page — use band changes to mark chapters. 8px spacing grid, 1200–1280px max content width, 24px (mobile) / 32–48px (desktop) gutters.

### Shape, depth & theme
Flat surfaces, hairline 1px borders in the muted colour, small radii (4–8px), almost no shadow. Read exact corner radius, border and shadow treatment from the screenshots. Light-first; for a dark mode use a near-black (not #000) background, ~90% white text, lift surfaces with lightness not shadow, re-check the accent.

## 6. Imagery plan

Measured from 9 section screenshots: mostly **typography with little or no imagery**, some photography (light, neutral, desaturated, Developer Tools). Still glance at the linked screenshots to match the exact mood.

### Slot 1 · 3. Content, 6. Content, 10. Content, 12. Content, 13. Content → illustration (uncertain)
- Light Developer Tools section with a few graphics — illustration, icon art or a product shot. Count: one consistent treatment across these sections. Aspect: 4:5. Look first: [ref 1](https://uidiscovery.com/media/captures/llamaindex/s-03.webp) [ref 2](https://uidiscovery.com/media/captures/llamaindex/s-06.webp) [ref 3](https://uidiscovery.com/media/captures/llamaindex/s-10.webp) [ref 4](https://uidiscovery.com/media/captures/llamaindex/s-12.webp)
- **Stock** — “technology flat illustration”: [Unsplash](https://unsplash.com/s/photos/technology-flat-illustration?orientation=portrait) · [Pexels](https://www.pexels.com/search/technology%20flat%20illustration/?orientation=portrait) · [Pixabay](https://pixabay.com/images/search/technology%20flat%20illustration/?orientation=vertical)
- **Generate** — `Flat vector illustration: abstract gradient mesh, friendly simplified shapes, limited palette of #ffffff, #000000, #fafafa plus one soft neutral, plenty of empty background, consistent line weight, neutral, desaturated mood. Aspect ratio 4:5. No text, no letters, no logos, no watermarks, no brand marks or trademarked products, no recognisable real people.`
- Uncertain: this could equally be photography, a product shot or icon art — open the reference screenshots and match what you see.

### Slot 2 · 16. Content → photography
- Photography, seen in the capture (people, adult). (measured from the screenshot) Count: a set of ~6 in one consistent style. Aspect: 4:5. Look first: [ref 1](https://uidiscovery.com/media/captures/llamaindex/s-16.webp)
- **Stock** — “abstract gradient mesh wide shot”: [Unsplash](https://unsplash.com/s/photos/abstract-gradient-mesh-wide-shot?orientation=portrait) · [Pexels](https://www.pexels.com/search/abstract%20gradient%20mesh%20wide%20shot/?orientation=portrait) · [Pixabay](https://pixabay.com/images/search/abstract%20gradient%20mesh%20wide%20shot/?orientation=vertical)
- **Generate** — `Photograph: abstract gradient mesh wide shot. Composition: simple composition with one focal point. Lighting: soft natural daylight, airy and high-key. Neutral, desaturated colour grade that sits on #f5f5f5, #000000, #fafafa. Photorealistic, natural textures, shallow depth of field, 35mm lens. Aspect ratio 4:5. No text, no letters, no logos, no watermarks, no brand marks or trademarked products, no recognisable real people.`

- **Logos** — 1. Logo cloud: Never generate or stock-source logos. Use your real customers'/partners' marks with permission (monochrome, equal height), or neutral wordmark placeholders until you have them. Integration icons: each tool's official brand kit (Simple Icons is a quick start; trademarks still apply).
- **No imagery (type-led)** — 2. Content, 4. Stats, 5. Statement, 7. Features, 8. Features, 9. Statement, 11. Features, 14. Features, 15. Content: No imagery needed — type, spacing and colour do the work. If the screenshot shows a background graphic, rebuild it in CSS (gradient, grid, noise) or a short muted video, not a stock photo.

### If the screenshots show a different style
Open the refs first and name what you actually see (photo / product UI / illustration / abstract gradient or 3D / type only). If it differs from the guess, switch routes — UI is built in code, type-only needs no image, otherwise:
- **abstract 3D / gradient** — generate `Abstract 3D render: soft, matte clay shapes and a smooth gradient in #7f7f7f and #fafafa on a #ffffff background. Lighting: soft natural daylight, airy and high-key; neutral, desaturated tone. Minimal and calm, generous negative space on one side for text. Aspect ratio 16:9. No text, no letters, no logos, no watermarks, no brand marks or trademarked products, no recognisable real people.` · stock: [“abstract gradient 3d” on Unsplash](https://unsplash.com/s/photos/abstract-gradient-3d?orientation=landscape) (same query works on Pexels/Pixabay)

- **Licences:** Unsplash and Pexels images are free for commercial use under their own licences; Pixabay under the Pixabay Content License. Check each image's page before use, credit the photographer when asked (good practice anyway), and skip images showing logos, trademarks, branded products or identifiable people without a release.
- **AI generation:** the prompts work with any image model (GPT Image, Imagen, Flux, Midjourney, Firefly, SDXL…) — check its commercial-use terms. Never put the reference brand's name or “in the style of <brand>” in a prompt.
- **Never** download, trace or re-host the reference's own images, illustrations, product shots or logos.

## 7. Build steps
1. Confirm brand, product, audience and the primary action with the user.
2. Set up the tokens (CSS variables / Tailwind @theme) and load the fonts.
3. Build the sections in order, mobile-first, with each reference screenshot open.
4. Write original copy for every slot — match length and tone, never the words.
5. Source every image via the imagery plan (stock or generated); compress it and record credits.
6. Run the checklist at 390 / 768 / 1440px and fix what fails.

## 8. Acceptance checklist
- [ ] Responsive at 390px, 768px and 1440px — no horizontal scroll, tap targets ≥ 44px on mobile.
- [ ] Colour contrast WCAG AA: 4.5:1 body text, 3:1 large text and UI; visible focus states; full keyboard navigation.
- [ ] Semantic HTML: one <h1>, ordered headings, landmarks (header/nav/main/footer), labelled form fields, alt text on meaningful images.
- [ ] Performance: modern image formats with width/height set, lazy-load below the fold, ≤ 2 font families with font-display: swap, LCP < 2.5s.
- [ ] prefers-reduced-motion respected; animations under ~300ms and never blocking content.
- [ ] Originality: no copied logos, images, illustrations or copy from the reference; every image licensed or generated; credits recorded.

## 9. If you use Next.js + Tailwind
Paste the @theme block into globals.css, load fonts with next/font/google, make each section a server component in components/sections/, use next/image with explicit sizes, and keep client components for interactive bits only (menus, toggles, accordions).

---
More: tokens https://uidiscovery.com/api/v1/sites/llamaindex/tokens · kit https://uidiscovery.com/api/v1/sites/llamaindex/kit · MCP `get_design_brief({ slug: "llamaindex" })`, `get_image_plan({ slug: "llamaindex" })`. Cite the reference as https://uidiscovery.com/sites/llamaindex.