The Financial OS for AI Agents — an x402 invoice middleware platform built, operated, and evolved entirely by a multi-agent AI company.
Invoica lets AI agents create, send, and settle invoices autonomously using the x402 payment protocol. It is a full-stack SaaS platform — REST API, TypeScript SDK, and a Next.js dashboard — that handles:
- Invoice creation & management with full lifecycle (draft → sent → settled)
- On-chain settlement detection across Base, Polygon, and Solana (in progress)
- x402 paywall middleware — any API endpoint can require a USDC micropayment
- Agent wallet infrastructure — agents hold USDC wallets and spend autonomously per LLM call
- Tax compliance — automated US, EU, and Japan regulatory monitoring
Invoica is built by an AI company. Every line of application code was written by AI agents. The humans set strategy; the agents execute:
You (Owner) → CEO bot (Telegram) → sprint JSON → agents build it → deployed to production
Since Week 1, 1,017 git commits have been made by 26 agents across 78 sprints — zero human-written application code.
┌─────────────────────────────────────────────────────────────────┐
│ Owner (You) │
│ Go/No-Go · Budget · Final approval │
└────────────────────────┬────────────────────────────────────────┘
│ Telegram
┌────────────────────────▼────────────────────────────────────────┐
│ CEO Agent (Claude Sonnet) │
│ Vision · Sprint planning · Conflict arbitration │
│ Wallet monitoring · Strategic pivots │
└──┬──────────┬──────────┬──────────┬────────────────────────────┘
│ │ │ │
┌──▼──┐ ┌───▼──┐ ┌────▼───┐ ┌───▼──────┐ ┌─────────┐
│ CTO │ │ CMO │ │ CFO │ │ BizDev │ │Supervisor│
│Mini │ │Manus │ │MiniMax │ │ MiniMax │ │ Claude │
│ Max │ │ AI │ │ │ │ │ │+ Codex │
└──┬──┘ └───┬──┘ └────────┘ └──────────┘ └─────────┘
│ │
│ Weekly content plan → X/Twitter (@invoica_ai)
│
┌──▼───────────────────────────────────────────────────────────┐
│ Coding Agents (MiniMax M2.5) │
│ backend-core · backend-ledger · backend-tax │
│ frontend · devops · security │
└──────────────────────────────────────────────────────────────┘
┌───────────────────────────────────────────────────────────────┐
│ Support / Watchdog Agents │
│ supervisor · execution-verifier · execution-watchdog │
│ pipeline-health-monitor · conflict-analyzer │
│ sprint-retrospective · test-runner · test-failure-predictor │
│ test-utility-generator · market-intelligence · skills │
│ conway-integration · memory-agent · telegram-support │
└───────────────────────────────────────────────────────────────┘
Every sprint runs without human involvement once the sprint JSON is queued:
CEO writes sprints/week-N.json
↓
sprint-runner (every 30 min)
↓
orchestrate-agents-v2.ts
├─ MiniMax M2.5 → writes code (one file per API call)
├─ Claude Sonnet → Supervisor 1 (architecture, security, tests)
├─ OpenAI Codex → Supervisor 2 (parallel review)
└─ Claude Sonnet → CEO arbiter (resolves supervisor conflicts)
↓ all tasks done
post-sprint-pipeline.ts
├─ Jest test suite
├─ CTO (MiniMax) → deploy or create bug-fix sprint
├─ git push → github.com (triggers CI)
├─ GitHub Actions → TypeScript check + ecosystem verify
└─ Vercel → app.invoica.ai
↓
Telegram alert to Owner
Dual supervisor consensus:
- Both approve → task merged (averaged score)
- Both reject → task retried with feedback
- Conflict → CEO reads both reviews and makes the call
Result: 593 of 639 tasks approved (92.8%) across 78 sprints.
Invoica uses the x402 protocol for AI-native micropayments. Agents hold real USDC wallets on Base and pay each other for LLM compute:
Coding agent wallet ──pays 0.001 USDC──► x402 gateway ──► CTO wallet (seller)
- Every LLM inference call requires a USDC micropayment
- Agent wallets monitored in real-time by CEO bot; alerts fire on low balance
- Backend exposes
/v1/ai-inferencebehind a 402 paywall - Multi-chain expansion in progress: Base ✅ · Polygon (Sprint 10) · Solana (Sprint 12)
| Service | Port | What it does |
|---|---|---|
| backend | 3001 | Express API — invoices, settlements, x402, API keys, webhooks |
| openclaw-gateway | 18789 | AI agent execution gateway — routes LLM calls, manages sandboxes |
| ceo-ai-bot | — | Telegram bot — 24/7 CEO interface, wallet monitoring, tool execution |
| mission-control | 3010 | Agent ops dashboard — 26 agents, tasks, logs, audit trail |
| Agent | Schedule | Role |
|---|---|---|
| sprint-runner | every 30 min | Executes pending sprint tasks via dual-supervisor orchestrator |
| git-autodeploy | every 5 min | git pull on new commits, self-heals crashed processes |
| cto-email-support | every 5 min | Handles inbound support emails autonomously |
| ceo-review | every 2 hrs | CEO reviews CTO proposals and approves/rejects backlog items |
| heartbeat | every hour | Writes health.json with real system state for CEO context |
| memory-agent | every hour | Daily logs + long-term memory (persists outside app dir) |
| x-admin-post | every 30 min | Posts approved content to @invoica_ai |
| cmo-daily-watch | 08:00 UTC daily | Market intelligence and competitor scan |
| docs-generator | 04:00 UTC daily | Auto-updates API docs from git log + routes, then deploys |
| cto-daily-scan | 09:00 UTC daily | Tech-watch: GitHub releases, HN, x402 ecosystem |
| cmo-weekly-content-plan | Sun 06:00 UTC | Generates full week of X posts (CTO + CEO review before posting) |
| bizdev-weekly | Sun 06:00 UTC | Scans for partnership and growth opportunities |
| cfo-weekly | Mon 07:00 UTC | Financial report: burn rate, runway, revenue |
| tax-watchdog-us | Mon 07:00 UTC | US crypto payment tax regulation monitoring |
| tax-watchdog-eu-japan | Mon 08:00 UTC | EU + Japan regulatory monitoring |
├── agents/ # 26 AI agent configs (agent.yaml + prompt.md)
├── backend/
│ └── src/
│ ├── routes/ # invoices, settlements, api-keys, ai-inference, ledger, webhooks
│ ├── services/ # tax, ledger, api-keys, settlement detection (EVM + Solana)
│ └── middleware/ # auth, rate-limiting, x402 paywall
├── frontend/ # Next.js dashboard → app.invoica.ai
├── sdk/ # TypeScript SDK for programmatic access
├── scripts/
│ ├── orchestrate-agents-v2.ts # Dual-supervisor sprint orchestrator
│ ├── post-sprint-pipeline.ts # Tests → CTO review → git push → Vercel deploy
│ ├── sprint-runner.ts # PM2 cron — picks up pending sprints every 30 min
│ ├── git-autodeploy.sh # Server self-deployment + process self-healing
│ └── ... # 15+ more autonomous agent scripts
├── sprints/ # 78 sprint JSON files (week-2 through week-74 + bugfixes)
├── plans/ # CEO strategic plans and March roadmap
├── reports/ # Auto-generated: CEO, CTO, CMO, CFO, BizDev reports
├── memory/ # Long-term agent memory (outside app dir, survives git wipes)
└── docs/ # Architecture, API contract, ADRs, learnings
| Layer | Technology |
|---|---|
| Coding Agents | MiniMax M2.5 |
| Executive Agents | Claude Sonnet (CEO, Supervisor 1, CEO arbiter) |
| Secondary Review | OpenAI o4-mini (Codex — Supervisor 2) |
| CMO / Content | Manus AI |
| Backend | Node.js, Express, TypeScript, Supabase |
| Frontend | Next.js 16, React, Tailwind CSS |
| SDK | TypeScript |
| Payments | x402 protocol, USDC on Base (Polygon + Solana in progress) |
| Agent Gateway | OpenClaw |
| Infrastructure | Hetzner VPS, PM2, GitHub Actions, Vercel |
| Monitoring | Mission Control (self-hosted), Telegram, healthchecks.io |
POST /v1/invoices Create invoice
GET /v1/invoices List invoices (paginated)
GET /v1/invoices/:id Get invoice by ID
GET /v1/invoices/number/:n Get invoice by number
POST /v1/api-keys Create API key
GET /v1/api-keys List API keys
POST /v1/api-keys/:id/rotate Rotate API key
GET /v1/settlements Settlement history
GET /v1/ledger Ledger entries
POST /v1/webhooks Register webhook
GET /v1/ai-inference AI inference (x402 paywall — requires USDC)
GET /health System health
The company autonomously adjusts operations based on revenue (defined in tier.json):
| Tier | MRR Threshold | Active Agents | Behaviour |
|---|---|---|---|
| Pre-launch | — | 26 | Building + beta user acquisition |
| Normal | $5,000+ | 18 | Full capabilities, replication mode on |
| Low Compute | $2,000+ | 15 | Pause non-revenue features, cheaper models |
| Critical | $500+ | 8 | Revenue recovery only |
| Dead | $0 | 2 | CEO + CTO only, human intervention required |
| Metric | Value |
|---|---|
| Sprints completed | 78 |
| Tasks approved | 593 / 639 |
| Approval rate | 92.8% |
| Total git commits | 1,017 |
| AI agents | 26 (6 coding + 20 executive/support/watchdog) |
| Always-on processes | 4 |
| Cron agents | 15 |
| Beta launch | February 27, 2026 |
| Monthly burn | ~$60–95/month (LLM API costs) |
| Agent treasury | $102.96 USDC across 8 wallets |
MIT
Built by 26 AI agents. Reviewed by Claude + Codex. Operated on Hetzner. 593 tasks approved. 1,017 commits. Zero human-written application code.