How the autonomous AI infrastructure behind Agents-as-a-Service works — from the development engine to the self-monitoring supervisor that runs 24/7.
Last updated: 2026-03-19
We built an AI platform called AaaS (Agents-as-a-Service) where businesses email tasks to specialist AI agents and get work done — no app required. Behind the visible websites, there's an autonomous infrastructure layer called AntiGravity that builds features, monitors health, and improves the system continuously without human intervention.
Think of it as: the websites are the storefront, and AntiGravity is the factory, warehouse manager, and quality inspector — all running autonomously.
Each domain serves a specific purpose in the customer journey:
From first visit to active user, everything is automated. No human intervention required except for billing setup.
Key insight: The entire flow — from signup to first task — runs without any human involvement. The context engineering pipeline uses Google's Vertex AI (Gemini 2.0 Flash) to understand each business and personalize the experience in under 60 seconds.
AntiGravity is the development and operations engine. It consists of three components:
A template system for spinning up AI-powered projects. Every AaaS project is born from the same base — ensuring consistent quality, security, and operational DNA across everything we build.
| Layer | What It Does | Example |
|---|---|---|
| 1. Directives | Operating procedures (SOPs) | "Always validate inputs before processing" |
| 2. Context | Project-specific memory | Brand guidelines, sprint goals, KPIs |
| 3. Knowledge | Best practices library | 4,200+ reusable AI skills in the Vault |
| 4. Security | Guardrails and safety rules | "Never expose API keys, always sanitize HTML" |
| 5. Orchestration | AI decision-making logic | Which model to use, when to escalate |
| 6. Execution | Deterministic scripts | Health checks, deployments, reports |
An autonomous monitoring system that watches over the entire AaaS ecosystem. It runs on a schedule — no human needed to start it.
| Schedule | Scan | What It Checks | Output |
|---|---|---|---|
| Every 4 hours | Infrastructure Pulse | All domains up? Functions running? Content intact? (No SupraForge overwrites) | Alert if anything fails |
| Daily 08:47 | Operations Brief | New signups, email deliveries, pipeline completions, error rates, Firestore stats | Daily Report → |
| Daily 10:30 | Content Generator | Which blog collections are thin? Auto-generates 5 new entity drafts | Blog submissions |
| Monday 08:00 | Context Sync | Re-indexes all 5,865 skills and directives into Pinecone for semantic search | Sync log |
| Monday 09:00 | Weekly Review | Cross-project infrastructure audit, stale indexes, deployment drift | Weekly Report → |
| Wednesday 14:23 | Strategy Review | Growth metrics, market positioning, conversion analysis, build priorities | Strategy Report → |
This means: Every morning at 08:47, a report is generated showing what happened overnight — signups, emails sent, pipeline completions. Every Monday, a deeper audit runs. Every Wednesday, strategic analysis. All deployed as branded HTML pages at aaas-design.web.app.
A repository of 4,200+ reusable AI capabilities — from Next.js patterns to TDD workflows to security scanning. When the system needs to build something new, it searches the Vault first (semantic search via Pinecone) rather than writing code from scratch. This dramatically reduces development time and errors.
The content pipeline runs continuously, feeding the blog and driving SEO traffic:
This means: content creates itself → attracts users → converts them → onboards them → activates them. The entire funnel is autonomous.
Email is the primary interface for AaaS — both for activation and for the actual agent service.
Each email is composed by AI using the user's context package. Not templates — personalized content generated by Vertex AI based on the user's industry, tech stack, and company stage. Delivered via Resend with RFC 8058 unsubscribe headers.
The next major feature: users email agent-function@aaas.select → the system parses the task → routes to the right agent → executes → replies with results. No app needed — just email.
| Feature | Status |
|---|---|
| Delivery tracking (delivered, opened, clicked) | Live |
| Bounce handling (auto-disable after 3 bounces) | Live |
| Complaint handling (auto-unsubscribe) | Live |
| One-click unsubscribe (RFC 8058) | Live |
| Email preferences page | Live |
| Resend webhook integration | Needs Config |
The system is designed for maximum autonomy. Here's what runs itself and what still needs a human:
| User signup → context engineering → onboarding | End-to-end automated |
| 7-stage email activation drip | AI-personalized, scheduled delivery |
| Infrastructure health monitoring | Every 4 hours, auto-alerts on failure |
| Daily/weekly/strategic reports | Auto-generated and deployed |
| Blog content generation | GravityClaw → Factory → Blog pipeline |
| New feature development | AI builds, tests, deploys autonomously |
| Subscriber → platform bridging | Blog readers auto-converted to users |
| Azure AD app registration | Microsoft auth provider for enterprise users |
| DNS record management | SPF/DMARC email authentication records |
| Stripe payment configuration | Subscription billing and pricing tiers |
| API key rotation | Resend, Porkbun, and other external services |
| Strategic business decisions | Pricing, partnerships, feature prioritization |
Current open issues requiring human action: 7 items tracked here →
The infrastructure is designed for minimal cost at current scale:
| Service | Monthly Cost | Notes |
|---|---|---|
| Firebase (Spark/Blaze plan) | ~$0–25 | Pay-as-you-go, free tier covers most usage |
| Cloud Run (blog) | ~$0–10 | Only charges for actual requests |
| Vertex AI (Gemini 2.0 Flash) | ~$1–5 | Context pipeline: ~$0.001 per user signup |
| Resend (email) | Free / $20 | Free tier: 3,000 emails/month |
| Pinecone | Free / $70 | Free tier: 100K vectors sufficient for now |
| Domains (3) | ~$30/year | aaas.name, agents-as-a-service.com, aaas.blog |
| Claude Code (development) | $100–200 | AI development agent for autonomous building |
| Total (current) | ~$150–300/month | Scales with usage, not upfront |
A full 42-item improvement plan exists at aaas-design.web.app/improvement-plan.html. The highest-impact items:
| Resource | URL |
|---|---|
| Main Website | agents-as-a-service.com |
| Login / Sign Up | aaas.name |
| Knowledge Index | aaas.blog |
| All Deliverables | aaas-design.web.app/deliveries |
| Open Issues (Human Action) | aaas-design.web.app/open-issues |
| Improvement Plan (42 items) | aaas-design.web.app/improvement-plan |
| Daily Operations Report | aaas-design.web.app/daily-report |
| Strategy Report | aaas-design.web.app/strategy-report |
| Firebase Console | Firebase Console → |
| GitHub Repositories | github.com/ibossyNr1 |