OpenLen

Beautiful landing pages. AI-built. Open source. Lovable quality. Your code. Your subdomain. $19 / month.

---

What this is

OpenLen is an open-source AI landing page generator. Describe what you want; we produce a complete, conversion-validated landing page in under 60 seconds — pure HTML + Tailwind, no platform lock-in, no npm install on the output. Generated pages publish to <your-name>.openlen.com with one click; you can also export the HTML and host it anywhere.

Try it live at openlen.com. First user-published landing: inari.openlen.com.

Key differentiators:

• Quality gates built in. Every output passes six validation gates (a11y · conversion · mobile · SEO + brief-fidelity · security · performance) before delivery. No other AI page builder does this.
• Code you own. Single HTML file + Tailwind via CDN. Deploy to Vercel, Netlify, Cloudflare, GitHub Pages, your own server — anywhere static hosting works.
• 10× cheaper than Lovable. Smart routing across 7+ models on Google Gemini keeps real cost at ~$0.13/generation (measured across 5 representative briefs — see EVAL_PHASE_2.md).
• Slot-filling architecture. The AI picks block IDs from a curated 15-block library and fills slot JSON. Code assembles HTML deterministically. Bug loops (Lovable's #1 user complaint) are structurally impossible.
• Open source from day one. AGPL v3. Self-host, fork, modify.

Quick start — hosted (when public)

Visit openlen.com → describe your landing → your generated HTML downloads.

Quick start — self-host

git clone https://github.com/orbita-pos/openlen
cd openlen
npm install
cp .env.local.example .env.local
# Add GEMINI_API_KEY, DATABASE_URL (Neon), NEXTAUTH_SECRET
# Or leave MOCK_MODE=1 for offline dev (no DB/API needed)
npm run dev

For self-hosted production deployment (Hetzner box + nginx wildcard + Let's Encrypt + systemd), see infra/SETUP.md.

Then in another terminal:

curl -N -X POST http://localhost:3000/api/generate \
  -H "Content-Type: application/json" \
  -d '{"brief":"Landing page for FlowDeck, a Kanban tool for designers. Pricing: Free, Pro $29/mo, Team $99/mo."}'

The endpoint streams Server-Sent Events: progress events per pipeline step, then a final result event with the HTML, CSS, image URLs, cost breakdown, gate verdicts, and witness path.

Environment variables

Variable	Default	Purpose
GEMINI_API_KEY	none	Required when MOCK_MODE is off. Get one at ai.google.dev.
MOCK_MODE	off	Set to 1 to use canned responses (no API spend, no key needed).
OPENLEN_DOMAIN	none	Optional. Canonical URL used in generated meta tags.
INARIWATCH_DSN	none	Optional. Error monitoring DSN; auto-local in dev when blank.

See .env.local.example for the full list.

Architecture

brief
  ↓  classify           LFM2-24B-A2B          ($0.03 / $0.12 per M)
intent
  ↓  plan + factsLedger Kimi-K2.6             ($1.20 / $4.50)
plan + ledger
  ├─  fill (parallel)   Qwen3-235B-tput       ($0.20 / $0.60)  ← slot JSON only
  └─  images            FLUX.2-pro / -flex    ($0.03 / image)
filled blocks + images
  ↓  assemble (deterministic React SSR — NO LLM)
html
  ↓  6 quality gates    (parallel)
    a11y       (axe-core)
    conversion (Kimi K2.6 judge + banned-phrase regex)
    mobile     (360 px puppeteer + AI judge)
    seo+aeo    (cheerio + brief-fidelity regex)
    security   (regex + ESLint security rules)
    performance(cheerio bundle/lazy-load)
  ↓  refine (only if critical violations remain — max 2 attempts)
LandingPage { html, css, images, meta, cost, witnessPath, qualityGrade }

Why this beats single-shot

• Section-parallel — fill + image generation overlap. Wall time ~42 s on a typical 6-block brief vs minutes for single-shot competitors.
• Bug loops impossible — there is no markup-generating model in the pipeline. The AI never writes JSX or HTML. Lovable's #1 user complaint ("the AI keeps breaking my page") can't happen here by construction.
• Per-section iteration — /api/regenerate-section re-runs the fill step for one block, splices it back in, re-assembles. ~$0.001 per regeneration.
• Cheap models do 90 % of the work — expensive models gate only the critical paths.

Quality gates — the open lane

None of the major AI page builders (Lovable, Bolt, v0, Framer AI, Webflow AI) ship explicit gates. They rely on implicit model quality. OpenLen enforces:

1. A11y — axe-core: alt text, labels, WCAG AA contrast, heading hierarchy
2. Conversion — banned-phrase regex + AI judge for primary CTA / hero outcome language / placeholder detection
3. Mobile — 360 px Puppeteer snapshot, no horizontal scroll, tap targets ≥ 44 px
4. SEO + AEO + brief-fidelity — single H1, meta description, OG tags, schema.org JSON-LD, brief fact preservation (prices, dates, named people)
5. Security — 19 regex patterns + 17 ESLint security rules
6. Performance — bundle size budget, lazy-load enforcement

Pages that fail critical violations get one targeted refine pass. If they still fail, they ship with qualityGrade: needs_review — no surprises.

See INARI_DESIGN_ENGINE.md § 6 for the full design.

Project layout

inari-pages/
├── app/
│   ├── api/generate/route.ts        # POST endpoint (SSE streaming)
│   ├── api/regenerate-section/      # one-block regen endpoint
│   ├── layout.tsx
│   └── page.tsx                     # marketing landing
├── components/marketing/            # marketing site components
├── lib/
│   ├── orchestrator/
│   │   ├── index.ts                 # generateLandingPage() main
│   │   ├── classify.ts plan.ts fill.ts images.ts assemble.tsx refine.ts
│   │   ├── routing.ts               # central routing table + fallbacks
│   │   ├── master-prompt.ts         # high-level system prompt
│   │   ├── design-tokens.ts         # 5 palettes (mono-dark, indigo, emerald, ...)
│   │   ├── few-shots/               # hand-crafted reference HTMLs
│   │   └── types.ts                 # Zod schemas + TS types
│   ├── blocks/                      # 15 vendored Tailark/Magic UI/shadcn blocks
│   │   ├── hero/      features/     pricing/     testimonials/
│   │   ├── faq/       cta/          footer/
│   │   ├── _registry.ts             # block ID → component map
│   │   └── types.ts
│   ├── gates/                       # 6 quality gates
│   │   ├── a11y.ts conversion.ts mobile.ts seo.ts security.ts performance.ts
│   │   ├── brief-fidelity.ts        # brief fact verification
│   │   └── _browser.ts              # shared puppeteer instance
│   ├── together/                    # SDK wrapper + pricing + mock dispatch
│   ├── witness/                     # JSONL audit trail per generation
│   └── budget.ts                    # cost cap + per-step accounting
├── evals/                           # eval harness + per-brief outputs
├── LICENSES/                        # vendored block-library MIT licenses
└── LICENSE                          # AGPL-3.0 (this repo's code)

Eval

Five representative briefs run end-to-end against real Gemini calls. Per-brief outputs, witness JSONL, cost breakdowns, and honest scoring live in evals/ and EVAL_PHASE_2.md:

Brief	Quality	Cost	Wall	Refines
01-saas-launch	4.5 / 5	$0.0812	50.1 s	0
02-portfolio	5 / 5	$0.1376	41.6 s	0
03-event-conference	5 / 5	$0.0522	44.9 s	0
04-ecommerce	5 / 5	$0.2201	44.3 s	0
05-agency	4.5 / 5	$0.1382	29.0 s	0

Averages: $0.126/gen · 42.0 s · 4.8/5 quality · 100 % gates pass first try.

To re-run:

MOCK_MODE=1 npm run eval        # offline pass against mocks
GEMINI_API_KEY=… npm run eval # real run, ~$0.50–1.40 total

Witness recordings

Every generation writes a JSONL audit trail to recordings/<generationId>.jsonl. Each line is a fully-typed WitnessRecord:

{"ts":"2026-05-15T18:42:11.213Z","generationId":"...","step":"classify",
 "decision":{"model":"lfm2-24b-a2b","reason":"Cheap classifier — small structured I/O.","isFallback":false,"fallbackChain":[]},
 "inputTokens":78,"outputTokens":52,"latencyMs":98,"costUsd":0.0000086,"mocked":false}

These are the audit trail for any output — open the path referenced by page.witnessPath and you see exactly which model produced what, why, at what cost, and on which fallback.

Roadmap

Shipped

• Sessions 1–7 — slot-filling pipeline + 6 quality gates + 5-brief eval (4.8/5 avg, $0.126/gen)
• Session 8 — sidebar slot editor (non-devs edit generated copy without prompting)
• Session 9 — image upload + cropping for hero/decorative slots
• Sessions 10 + 10.5 — Hetzner self-host deploy (nginx wildcard + Let's Encrypt wildcard cert + systemd + Kamal-less). Live at openlen.com.
• Session 11 — per-user subdomain publish flow. Click "Publish" → page lives at <sub>.openlen.com via atomic filesystem write + nginx wildcard. First user page: inari.openlen.com.
• Session 12 — in-iframe WYSIWYG text editing. Click any text in the preview, type, hit Enter — instant update, sidebar reflects, debounced reassemble persists to DB.

In progress (V3 pivot)

Architecture research (RESEARCH_FINDINGS.md) identified that the catalog approach has a ceiling. V3 pivots to:

• Curated design system (8 backgrounds + 20 palettes + 6 typography systems + 5 layout primitives) hand-tuned in claude.ai
• AI runtime composes from menu (Gemini planner + writer, single Google Gemini vendor, ~$0.015/gen — 8× cheaper than V1)
• Visual design knobs in the workspace — click thumbnails to swap bg/palette/typography/density/radius/decoration instantly. Zero AI call per knob change.
• No more AI image gen — replaced with Unsplash API + SVG decoration primitives + user uploads. Editorial photo quality, no FLUX cost or hallucinations.

See V3_AUDIT.md for the file-by-file pivot plan, SESSION_13_PROMPT.md and SESSION_14_PROMPT.md for the implementation sessions. Both pipelines coexist behind feature flag OPENLEN_PIPELINE_V3 during the rollout window.

After V3

• Custom domains (CNAME mydomain.com → openlen.com via dynamic certbot)
• Multi-page support (linked landings, shared design system)
• Per-tier image upload caps (free 5/page, pro unlimited)
• Stripe billing
• Show HN launch

Contributing

PRs welcome. AGPL v3 means your modifications stay open-source — including hosted modifications (the "network use" clause is the whole point of AGPL over GPL). By submitting code you agree to license it under the same terms.

Issues for bug reports, feature requests, and brief-fidelity false-positives all go to github.com/orbita-pos/openlen/issues.

License

AGPL-3.0-only for the codebase. Vendored block libraries (Tailark, shadcn/ui, Magic UI, HyperUI) keep their original MIT licenses — see LICENSES/.