System Architecture

Relevant source files

The following files were used as context for generating this wiki page:

• CLAUDE.md
• AGENTS.md
• INFRASTRUCTURE_SETUP.md
• HERMES_FBIS_AUDIT.txt
• src/utils/template-router.js
• scripts/setup-infrastructure.js
• gen-lp.md

The FusionOps (internal name LP Factory) system is an Astro-based control plane designed for the creation, deployment, and tracking of PPC and affiliate marketing landing pages. The architecture follows a monorepo structure consisting of a frontend dashboard, multiple specialized Cloudflare Workers for backend logic and event tracking, and an integrated autonomous agent system (Hermes/FBIS) for risk analysis and operations.

Sources: CLAUDE.md:10-15, AGENTS.md:3-5

Core Monorepo Components

The system is organized into several distinct applications and services that handle different parts of the landing page lifecycle.

Monorepo Structure

App	Technology	Purpose
apps/lander	Astro + React	Main UI dashboard for campaign and template management.
apps/api-worker	Cloudflare Worker + D1	Primary CRUD API for campaigns, sites, and templates.
apps/pixel-worker	Cloudflare Worker + D1	First-party tracking pixel for clicks and conversions.
apps/worker	Cloudflare Worker + D1	Webhook handling, lead notifications, and validation.
apps/fbis-mcp-server	Python FastMCP	Autonomous agent integration for ban intelligence.

Sources: CLAUDE.md:28-66, INFRASTRUCTURE_SETUP.md:71-76

Architectural Data Flow

The following diagram illustrates the high-level interaction between the frontend, the various Cloudflare Workers, and the underlying databases.

The system leverages Cloudflare's Edge network to ensure low-latency tracking and landing page delivery. Sources: INFRASTRUCTURE_SETUP.md:5-40, CLAUDE.md:51-70

Database Layer

The project utilizes a hybrid database approach combining Cloudflare D1 for edge-localized storage and Neon Postgres for centralized site configuration and settings.

D1 SQLite Databases

D1 instances are used where edge performance is critical, such as tracking and real-time API responses.

• fusionops-main: Stores campaigns, templates, and site metadata.
• fusionops-pixel: Stores raw event data (clicks, user UUIDs, timestamps).
• fusionops-callback: Stores lead data received from external postbacks.

Sources: INFRASTRUCTURE_SETUP.md:71-76, CLAUDE.md:68-73

Neon Postgres

Neon is used as the primary relational store for higher-level management data.

• settings: A key-value store for global application configuration.
• sites: JSONB storage for complex landing page configurations.
• deploy_history: Tracks deployment status across various targets (Cloudflare, Vercel, Netlify).

Sources: INFRASTRUCTURE_SETUP.md:144-165, scripts/setup-infrastructure.js:338-360

Template and Build System

FusionOps uses a "Build-Variant" approach to generate landing pages. Templates are typically designed as Astro components or static HTML, then processed to inject dynamic values and tracking scripts.

Template Generation Logic

The template-router.js handles the routing and rendering of different template formats, including legacy modules and newer "Loveable" (React+Vite) imports.

Tracking Injection

A core requirement is the "zero-GTM" tracking architecture. The build pipeline automatically injects three layers of tracking:

1. Layer 1: window.dataLayer push for form events (form_start, form_submit).
2. Layer 2: First-party pixel via navigator.sendBeacon to the Pixel Worker.
3. Layer 3: Conditional Voluum dtpCallback scripts.

Sources: gen-lp.md:65-75, src/utils/template-router.js:154-180

Hermes + FBIS Autonomous System

The system includes a suite of autonomous agents powered by a Python FastMCP server. These agents perform nightly analysis to detect "ban patterns" and risk levels for advertising accounts.

Agent Teams

• FBIS Team (Intelligence): Composed of agents like ARGUS (Proxy risk) and CHRONO (Ban timeline prediction).
• LP Builder Team: Automates the construction of landing pages using the template system.
• Verdict Agent: Aggregates scores from other agents to produce a final verdict_score (0-100) and triggers auto-pause rules if scores exceed thresholds (e.g., score ≥ 80).

Sources: HERMES_FBIS_AUDIT.txt:30-45, HERMES_FBIS_AUDIT.txt:74-85

Risk Scoring API

The API Worker exposes specific endpoints for the Hermes agents to store and retrieve analysis data.

Endpoint	Method	Action
/api/analysis/risk-score	POST	Writes calculated risk scores to D1.
/api/accounts	GET	Queries advertising account status for analysis.
/api/pixel-events	GET	Retrieves click/conversion data for quality analysis.

Sources: HERMES_FBIS_AUDIT.txt:115-125, HERMES_FBIS_AUDIT.txt:168-180

Infrastructure Management

Infrastructure is managed via script-based automation and environment variable configuration.

Deployment Pipeline

Deployments are handled primarily through GitHub Actions, which execute the following steps:

1. Trigger on push to main.
2. Build the Astro site dashboard.
3. Execute inject-tracking.mjs to prepare landing page assets.
4. Deploy Cloudflare Workers using Wrangler.
5. Deploy the dashboard to Cloudflare Pages.

Sources: CLAUDE.md:162-170, INFRASTRUCTURE_SETUP.md:200-210

Setup Automation

The setup-infrastructure.js script automates the creation of D1 databases and the execution of SQL migrations across the three required instances.

// Database configuration used by setup script
const CONFIG = {
  databases: {
    main: 'fusionops-main',
    pixel: 'fusionops-pixel',
    callback: 'fusionops-callback'
  },
  migrations: {
    main: ['0001_init.sql', '0002_deploy_history.sql', '0002_add_templates.sql'],
    // ... other migrations
  }
};

Sources: scripts/setup-infrastructure.js:28-40, scripts/setup-infrastructure.js:158-180

The FusionOps architecture represents a highly distributed and automated system, optimized for performance at the edge while maintaining a robust centralized control plane for complex marketing operations and risk management.