# RateRight — System Overview ## Architecture at a Glance ``` ┌─────────────────────────────────────────────────────────┐ │ PUBLIC FACING │ │ │ │ rivet.rateright.com.au — The App (Next.js) │ │ rateright.com.au — Landing Page (Flask/Fly) │ │ Growth Engine — CRM API (Railway) │ │ Phone AI — Voice Receptionist (VAPI) │ │ │ ├─────────────────────────────────────────────────────────┤ │ BACKEND │ │ │ │ Supabase — Database, Auth, RLS, Edge Functions │ │ Stripe — Payments ($50/hire) │ │ nginx — Reverse proxy, SSL termination │ │ │ ├─────────────────────────────────────────────────────────┤ │ AI AGENT FLEET (OpsMan Prototype) │ │ │ │ 8 Clawdbot gateways, each a specialist agent │ │ Coordinated via fleet-state.json + JSONL inboxes │ │ This fleet IS the OpsMan product demonstration │ │ │ └─────────────────────────────────────────────────────────┘ ``` --- ## 1. The App — RateRight Marketplace **Stack:** Next.js 15 (App Router) + TypeScript + Tailwind CSS + shadcn/ui **Database:** Supabase (PostgreSQL + Row Level Security) **Auth:** Supabase Auth (email/password + OAuth) **Payments:** Stripe ($50 AUD flat fee per hire) **Hosting:** DigitalOcean VPS (syd1, 2 vCPU, 8GB RAM) via systemd **Domain:** rivet.rateright.com.au (production) — planned move to rateright.com.au **Build:** `npm run build` → systemd service `rateright-app` on port 3000 ### Key Features - **Contractor flow:** Sign up → Post job (voice or form) → Browse matches → Hire → Pay $50 - **Worker flow:** Sign up → Build profile (voice or form) → Get matched → Accept work → Build ratings - **AI matching:** Workers ranked by relevance to job requirements - **Voice-to-job:** Contractors speak job descriptions, AI fills the structured form - **Voice profiles:** Workers fill out profiles by speaking naturally - **Onboarding packs:** Client-side PDF generation for contractor compliance docs - **Browse Workers:** Contractors can discover workers directly - **CEO Dashboard:** `/ceo` — business metrics and admin overview ### Database Schema (Supabase) - Users (auth) → Profiles (contractor or worker type) - Jobs (posted by contractors) - Applications / Matches - Ratings & Reviews - Payments (Stripe integration) - Phone numbers + names required on all signup paths (DB trigger writes from auth metadata) - RLS enforced across all tables ### Tech Details - **Port:** 3000 (behind nginx) - **Service:** `systemd: rateright-app` - **Deploy:** `cd /home/ccuser/the-50-dollar-app && npm run build && systemctl restart rateright-app` - **Workspace:** `/home/ccuser/the-50-dollar-app` --- ## 2. Growth Engine — CRM & Sales Intelligence **Stack:** Node.js + Express **Database:** Supabase (separate project from the app) **Hosting:** Railway (rateright-growth-production.up.railway.app) **Workspace:** `/home/ccuser/rateright-growth` ### API Endpoints (15+) - Leads CRUD (30+ fields per lead including AI analysis, contact history, pipeline tracking) - SMS sending and sequences (via Twilio — currently all deactivated) - Call logging with post-call AI analysis - Intel briefs (AI-generated business intelligence per lead) - Dashboard stats - Notion sync - Team management ### Integrations - Notion (business ops, planning) - Slack (team coordination) - Gmail (email) - Google Sheets (reporting) - Twilio (SMS + voice) - Pushover (notifications) - Apollo (contact enrichment) ### Key Feature Post-call AI analysis that splits intel into personal vs business files — automatically enriches lead profiles after every call. --- ## 3. Phone AI — Voice Receptionist **Platform:** VAPI **Model:** GPT-4o with custom RateRight personality **Assistant ID:** 63ac4ff9-e562-4c8e-9c19-c1f1c87ed034 **Purpose:** Qualify inbound leads, log to Growth Engine CRM **Phone:** Business line (+61 468 087 171) --- ## 4. Landing Page — rateright.com.au **Stack:** Flask (Python) **Hosting:** Fly.io (rateright-au) **Purpose:** Current public-facing landing page (will be replaced by Next.js app) --- ## 5. AI Agent Fleet — The OpsMan Prototype The 8-agent fleet runs the business AND demonstrates what OpsMan would be for contractors. ### The Fleet | Agent | Port | Model | Role | |-------|------|-------|------| | **Rivet** 🔧 | 18789 | Claude Opus 4.6 | Chief of Staff — coordinates all agents, strategy, decision prep | | **Builder** 🔨 | 18790 | Claude Opus 4.6 | Development — ALL code changes, testing, deployment | | **Susan** 💼 | 18792 | DeepSeek | Sales & Marketing — leads, CRM, outreach, pipeline | | **Harper** 📊 | 18796 | DeepSeek | Finance & Legal — grants, compliance, tax, BAS | | **Sentinel** 🛡️ | 18800 | DeepSeek | DevSecOps — monitoring, infrastructure, security | | **Radar** 🔭 | 18804 | DeepSeek | Intelligence — market research, competitor analysis, trends | | **Herald** 📡 | 18808 | Claude Opus 4.6 | Communications — content creation, external comms review | | **Cog** ⚙️ | 18812 | DeepSeek | Operations — routine tasks, health checks, data processing | ### How Agents Work - Each agent is a **Clawdbot gateway** (Node.js process) with its own port, model, and workspace - Managed via **systemd** (`clawdbot-{agent}` services) - Communicate via: - **Shared inboxes** (JSONL files for async messaging) - **fleet-state.json** (centralized status tracking) - **queue.json** (per-agent task queues) - **Agent bridge** (`node scripts/agent-bridge.js `) - **Gateway-to-gateway WebSocket** (direct RPC between gateways) - **Telegram** integration for human communication (Rocky & Rivet group, topic-based) ### Fleet Coordination Infrastructure - **Stall detector** (*/5 min) — detects unresponsive agents - **Buddy check** (*/30 min) — cross-agent health monitoring - **Work generator** (every 2h) — creates tasks for idle agents - **Context monitor** (*/10 min) — tracks agent context window usage - **Crash loop guard** (*/2 min) — catches and recovers crash-looping agents - **Session archiver** (3 AM daily) — cleans stale sessions - **Inbox curator** (*/30 min) — manages JSONL message flow ### Agent Workspaces ``` /home/ccuser/ ├── the-50-dollar-app/ # Builder workspace (Next.js app + fleet scripts) ├── rateright-growth/ │ └── rivet/ # Rivet workspace (coordination, memory, plans) ├── susan/ # Susan workspace (sales) ├── harper/ # Harper workspace (finance) ├── sentinel/ # Sentinel workspace (devops) ├── radar/ # Radar workspace (intelligence) ├── herald/ # Herald workspace (comms) ├── cog/ # Cog workspace (operations) └── shared/ # Shared scripts and fleet state ├── fleet-state.json └── scripts/ ├── fleet-cli.js ├── fleet-update.js ├── stall-detector.js ├── buddy-check.js ├── work-generator.js ├── context-monitor.js ├── inbox.js └── inbox-curator.js ``` --- ## 6. Infrastructure | Component | Details | |-----------|---------| | **VPS** | DigitalOcean syd1, 2 vCPU, 8GB RAM (134.199.153.159) | | **OS** | Ubuntu Linux 6.8.0-100-generic (x64) | | **Node** | v22.22.0 | | **Reverse Proxy** | nginx (SSL termination, port routing) | | **SSL** | Let's Encrypt (auto-renewed) | | **Process Manager** | systemd (all services) | | **DNS** | rivet.rateright.com.au → VPS, rateright.com.au → Fly.io | ### External Services | Service | Purpose | Status | |---------|---------|--------| | Supabase | Database + Auth | ✅ Connected | | Stripe | Payments | ✅ Configured ($50 AUD product) | | Anthropic | Claude Opus 4.6 (Rivet, Builder, Herald) | ✅ Working | | DeepSeek | Agent models (Susan, Sentinel, Radar, Cog) | ✅ Working | | OpenAI | Various AI tasks | ✅ Configured | | Twilio | SMS + Voice | ✅ Configured | | Railway | Growth Engine hosting | ✅ Deployed | | Fly.io | Landing page hosting | ✅ Deployed | | VAPI | Phone AI | ✅ Active | | Brave Search | Web search | ✅ Configured | | Apollo | Contact enrichment | ✅ Working | | Notion | Business ops | ✅ Connected | --- ## 7. Current System Health (Feb 18, 2026) **Overall: 62%** (per Builder's system audit) | Area | Score | Notes | |------|-------|-------| | Infrastructure | 95% | Rock solid — VPS, services, SSL all healthy | | Agent Fleet | 85% | All alive but monitoring data lags | | Communication | 45% | Inboxes exist but cross-agent coordination broken | | Automation | 80% | All 12 cron jobs running, work generation low | | External Services | 75% | MiniMax broken, Moonshot suspended | | Memory | 40% | 4 agents barely maintaining context | | Task Workflow | 35% | Tasks queued but none being claimed or worked | | Model Strategy | 70% | Working but 2/4 providers down | ### Top 3 Gaps 1. **Task lifecycle broken** — tasks sit in queues, agents don't claim them 2. **Communication loop broken** — no cross-agent work handoffs happening 3. **Memory inequality** — 4 agents are practically amnesiac (1-3 memory files) --- *System built by Michael (prompting) + Builder (coding) + Rivet (coordination).* *The agent fleet itself IS the OpsMan product prototype — proving AI can run business operations.*