AI landing generator

MIT · one prompt → one self-contained HTML landing in ~30–60 seconds.

▶ Open the live generator

Try it in the browser — enter a prompt, pick a style, generate, preview, download landing.zip.
Part of the AI-Factory ecosystem · GitHub · AI-Factory source

A fast, self-hosted page generator. Two LLM agents — Architect (structure + visual plan) → Developer (single HTML file) — plus 20 style presets. No admin panel, no database, no QA pipeline: ideal for MVPs, A/B copy, and “ship the landing today” workflows.

| | AI landing generator | AI-Factory | |---|-------------------|------------------| | Time | ~30–120 s | >30 min | | Output | One index.html (+ optional ZIP) | Full product + gates | | Best for | Landing hypothesis, marketing experiments | Production software |

API (hosted): POST https://magic-ai-factory.com/landing-page-generation/api/generate — JSON body; send Origin / Referer for the same host (not a GET link in the browser). Optional: "agent_to_website": true — see docs/AGENT-TO-WEBSITE.md.

npm · npx

npx aicom-landing "SaaS landing for an AI task manager for remote teams"
npx aicom-landing-serve   # web UI on http://127.0.0.1:3847/
npm install -g aicom-landing

Package: npmjs.com/package/aicom-landing · requires one LLM API key in .env or env vars (see Quick start).

Prompt → ready page (demo GIF)

Floor demo (AICOM_LANDING_DEMO_MODE=true) needs no API key — serves a canned example landing for marketing hosts and README capture.

Regenerate: npm run demo-gif (Playwright + demo mode).

Screenshots

Generator UI (local preview)

Type a product brief, pick a visual preset (or Auto), optionally enable Agent-To-Website (embedded demo chat on the page), generate, preview in-page, download landing.zip.

npm run serve
# → http://127.0.0.1:3847/

UI languages: en (default), ru, es — set AICOM_LANDING_UI_LOCALE=en in .env, or open ?lang=ru for a one-off switch. Sample prompts: docs/PROMPTS.md.

Example landings (generated with this repo)

Real outputs from the CLI (DeepSeek, May 2026). Each file is one HTML document with an embedded demo AI agent (chat open in screenshots) — open locally or host anywhere static.

1 · SaaS task manager · midnight-terminal

Prompt: SaaS landing for an AI task manager built for remote teams — calm, trustworthy, dark command-center UI with clear pricing.

→ Open docs/examples/saas-task-manager.html

2 · Fintech payments · luxe-gold-obsidian

Prompt: Fintech app for cross-border payments for freelancers — trust, speed, premium dark UI.

→ Open docs/examples/fintech-payments.html

3 · Home solar · sage-organic

Prompt: Green energy startup landing for home solar panels — warm, optimistic, organic wellness aesthetic.

→ Open docs/examples/green-solar.html

Regenerate or add your own:

node cli.mjs "Your product pitch" --style aurora-glass --out docs/examples/my-landing.html
npx aicom-landing --list-styles

More detail: docs/examples/README.md.

Quick start

Requirements: Node.js 18+ (native fetch). One LLM provider:

• ANTHROPIC_API_KEY, or
• DEEPSEEK_API_KEY (recommended if OpenAI is region-blocked), or
• OPENAI_API_KEY, or
• local Ollama (OLLAMA_HOST / OLLAMA_MODEL)

cd aicom-landing
cp .env.example .env          # add your API key
npm run serve                 # web UI on :3847

CLI (writes output/index.html by default):

npx aicom-landing "SaaS landing for an AI task manager for remote teams"
npx aicom-landing "Green energy startup" --style sage-organic --out ./dist/page.html
npx aicom-landing "Florist with on-page AI guide" --agent

Keys and AICOM_LANDING_UI_LOCALE load from .env at startup (.env overrides shell exports). Provider choice and models: LLM providers below.

LLM providers

There is no provider toggle in the UI — the server picks the first configured cloud key, otherwise Ollama. Routing lives in llm/provider.js.

Where API keys live

| Environment | File / location | |-------------|-----------------| | Local (npm run serve, CLI) | .env next to package.json (copy from .env.example; never commit — listed in .gitignore) | | Docker | Host ./.env, mounted via env_file: in docker-compose.yml | | Production (VPS) | e.g. /opt/aicom-landing/.env on the server — separate from your laptop; edit on the host, then docker compose up -d --build |

Keys are read only on the server (Architect → Developer pipeline). The browser never sees DEEPSEEK_API_KEY or other secrets.

After changing .env, restart the process (npm run serve or recreate the container).

Which provider is used (priority)

| Order | If this env var is set | Default model | Optional model override | |------:|------------------------|---------------|-------------------------| | 1 | ANTHROPIC_API_KEY | claude-sonnet-4-6 | ANTHROPIC_MODEL | | 2 | DEEPSEEK_API_KEY | deepseek-chat | DEEPSEEK_MODEL, DEEPSEEK_BASE_URL (default https://api.deepseek.com/v1) | | 3 | OPENAI_API_KEY | gpt-4o-mini | OPENAI_MODEL, OPENAI_BASE_URL (Groq / other OpenAI-compatible APIs) | | 4 | (none of the above) | Ollama | OLLAMA_HOST (default http://127.0.0.1:11434), OLLAMA_MODEL (default llama3.2) |

DeepSeek only: set DEEPSEEK_API_KEY and leave ANTHROPIC_API_KEY / OPENAI_API_KEY unset (or commented out) in .env.

Example .env (DeepSeek):

DEEPSEEK_API_KEY=sk-...
# DEEPSEEK_MODEL=deepseek-chat
# DEEPSEEK_BASE_URL=https://api.deepseek.com/v1
AICOM_LANDING_UI_LOCALE=ru

Generation tuning (optional)

| Variable | Default | Purpose | |----------|---------|---------| | AICOM_LANDING_JSON_RETRIES | 3 | Retries per stage when model JSON is malformed or truncated | | AICOM_LANDING_FETCH_RETRIES | 3 | Retries for transient LLM HTTP/network errors |

Architect / Developer output token caps are fixed in code (8192 / 16384). Troubleshooting JSON errors: docs/DEPLOY.md.

Configuration

| Variable | Default | Purpose | |----------|---------|---------| | AICOM_LANDING_HOST | 127.0.0.1 | Preview server bind | | AICOM_LANDING_PORT / PORT | 3847 | Preview server port | | AICOM_LANDING_UI_LOCALE | en | UI + fallback landing language (en | ru | es) | | AICOM_LANDING_BADGE_ENABLED | true | Footer “Powered by” pill on generated pages | | AICOM_LANDING_BADGE_URL | https://magic-ai-factory.com/ | Badge link | | AICOM_LANDING_BADGE_LABEL | Powered by AI-Factory | Badge text | | AICOM_LANDING_RATE_LIMIT | 20 | Max POST /api/generate per client IP per 15 minutes (0 = off) | | AICOM_LANDING_TRUST_PROXY | (unset) | Set to true only behind your own reverse proxy so rate limits honor X-Forwarded-For safely | | AICOM_LANDING_BASE_PATH | (empty) | Browser URL prefix when the app is not at domain root (e.g. /landing-page-generation). Prefer a subdomain instead — see docs/DEPLOY.md. | | AICOM_LANDING_AGENT_STRICT | (on) | Replace LLM agent widget if risky JS detected (false to disable) | | AICOM_LANDING_SKIP_AGENT_INJECT | (off) | true = never inject safe fallback widget |

Agent-To-Website: docs/AGENT-TO-WEBSITE.md — embedded chat, demo mode, security model.

Production / reverse proxy: see docs/DEPLOY.md (subdomain recommended, TLS, nginx).

The badge is injected after generation (lib/badgeConfig.mjs), not left to the LLM — so URL/label stay exact. When the page includes #aicom-agent (Agent-To-Website), the badge is placed left of the chat FAB (bottom: 0.625rem, right: 5.85rem) so it does not cover the button.

Generation limits (built-in): Architect uses up to 8192 output tokens; Developer up to 16384. Malformed or truncated JSON is retried (AICOM_LANDING_JSON_RETRIES, default 3). See docs/DEPLOY.md.

Security: the preview iframe uses sandbox (scripts/forms only; no allow-same-origin) so generated JS cannot read this UI’s cookies or DOM. Responses add CSP on the shell page and on /preview/:id. Still treat LLM output as untrusted: host behind auth/TLS on anything beyond a personal machine.

| Concern | Mitigation | |--------|------------| | Preview / ZIP payload | Session IDs are unguessable + TTL; CSP on preview; iframe sandbox isolates the parent UI | | CSRF on POST /api/generate | Origin / Referer hostname must match Host hostname (default ports like :443 / :80 ignored so TLS proxies work) | | Abuse / cost | Optional rate limit: AICOM_LANDING_RATE_LIMIT (default 20 POSTs per IP per 15 min; 0 disables) | | Agent widget JS | Strict audit + safe fallback; demo chat does not call your LLM — docs/AGENT-TO-WEBSITE.md |

Programmatic POST /api/generate from non-browser tools must send Origin or Referer whose hostname matches the request Host hostname (e.g. -H "Origin: http://127.0.0.1:3847").

Docker

Build once; pass secrets at run time only (never bake keys into the image). Put keys in .env on the host and pass --env-file .env (or use Compose below) so they survive container restart/recreate — see docs/DEPLOY.md.

The image sets AICOM_LANDING_HOST=0.0.0.0 so the process listens on all interfaces (unlike the default 127.0.0.1 when run with Node locally). There is no authentication and no HTTPS inside the container — expose it only on trusted networks or put TLS + auth (reverse proxy, VPN, etc.) in front. The process runs as a non-root user inside the image (uid 1001).

cp .env.example .env   # DEEPSEEK_API_KEY, AICOM_LANDING_UI_LOCALE=en, …

# Preview UI (Compose — env persists on host)
docker compose up -d --build

# Or one-off / manual
docker build -t aicom-landing .
docker run --rm -p 3847:3847 --env-file ./.env \
  --entrypoint node aicom-landing /app/preview-server.mjs

# CLI one-shot
docker run --rm --env-file ./.env -v "$PWD/out:/out" aicom-landing \
  "Fintech app for freelancers" --out /out/index.html

# Disable badge / custom label
docker run --rm -p 3847:3847 \
  -e DEEPSEEK_API_KEY \
  -e AICOM_LANDING_BADGE_ENABLED=false \
  --entrypoint node aicom-landing /app/preview-server.mjs

How it works

User prompt + style preset
        ↓
   Architect (JSON: layout, copy language, visual plan, sections)
        ↓
   Developer (single HTML: embedded CSS, inline SVG, optional JS)
        ↓
   Badge inject (env) → preview / ZIP / disk

• 20 presets in styles/presets.json (glass, brutalist, editorial, organic, cyberpunk, …).
• Landing copy language follows the prompt; if unclear, uses ui_locale from the UI (not English by default).
• Funnel: optional footer CTA to magic-ai-factory.com when the badge is enabled.

Community

• License (MIT)
• Code of conduct
• Contributing
• Security policy

Repo layout

aicom-landing/
├── cli.mjs                 # CLI entry
├── preview-server.mjs      # Web UI + /api/generate + preview + zip
├── public/index.html       # Generator UI
├── lib/
│   ├── generate.mjs        # Architect → Developer pipeline
│   ├── agentToWebsite.mjs  # Agent-To-Website widget + security audit
│   ├── badgeConfig.mjs     # Powered-by env + HTML inject
│   ├── uiLocale.mjs        # UI strings (en / ru / es)
│   └── …
├── llm/                    # provider + prompts
├── styles/presets.json
├── docs/
│   ├── AGENT-TO-WEBSITE.md # Embedded agent mode + security
│   ├── screenshots/        # README images
│   └── examples/           # Sample generated landings
└── Dockerfile

See Contributing for how to propose changes.

License

This project is released under the MIT License.