AI Control Tower

Personal dashboard for monitoring quota usage across multiple AI providers in real time — from a single interface, without switching tabs.

When you work with multiple AI providers simultaneously, quota management becomes a daily friction point. You hit a rate limit mid-session, switch to another model, forget which one was about to reset, and lose context. AI Control Tower solves this by exposing a unified live dashboard — provider health, quota progress bars, and low-quota alerts — polling every 10 seconds from a lightweight local server.

---

What It Monitors

Provider	What you see
Claude	Session quota (% used), 5h window reset time, optional API prepaid balance
GitHub Copilot	Session usage from github.com cookie
DeepSeek	API balance + optional monthly spend
Tavily	Search credit usage from session cookie or API key
Codex CLI	Auth token from ~/.codex/auth.json + optional API balance
Antigravity (Codeium)	Local language server process — no credential needed

Every connector reports a unified health status: OK · WARNING · CRITICAL · BLOCKED.

---

Architecture

Browser dashboard (SPA)
        │ polls every 10s
        ▼
Fastify API server (src/index.ts)
        │
        ▼
ConnectorFactory
(static registry — resolved by provider name at runtime)
        │
        ├── ClaudeConnector
        ├── CopilotConnector
        ├── DeepSeekConnector
        ├── TavilyConnector
        ├── CodexConnector
        └── AntigravityConnector
              │
              ▼
        BaseConnector
        (abstract + in-memory TTL cache)
        ↓
        fetchMetricsRaw() → ProviderMetrics

The dashboard is a dependency-free SPA (public/index.html + app.js + style.css). No framework, no build step — just fetch and render.

Unified metrics schema

Every connector returns the same shape:

interface ProviderMetrics {
  provider: string;
  status: 'active' | 'inactive' | 'error';
  health: 'OK' | 'WARNING' | 'CRITICAL' | 'BLOCKED';
  models: ModelMetrics[];
  resetAt: string | null;
  lastUpdatedAt: string;
}

interface ModelMetrics {
  modelId: string;
  modelName: string;
  quota: MetricQuota;
  resetAt: string | null;
}

interface MetricQuota {
  type: 'tokens' | 'requests' | 'credits' | 'percent' | 'currency';
  total: number;
  used: number;
  remaining: number;
}

Adding a new provider means implementing one method — fetchMetricsRaw() — and registering the connector. The dashboard and API pick it up automatically.

---

How Credentials Work

No credential is extracted, generated, or shared. The connectors reuse what is already present on your machine after a normal browser login — the same cookies and tokens your browser stores, the same way a DevTools session or browser extension would.

To retrieve session credentials:

1. Log in to the provider's website
2. Open DevTools → Network → reload
3. Click any authenticated request → copy the Cookie or Authorization header

Every variable in .env is optional — leaving one blank causes that connector to report inactive instead of erroring.

---

API

GET /v1/providers

Health summary for all registered providers:

{
  "claude": "OK",
  "copilot": "WARNING",
  "deepseek": "OK",
  "tavily": "OK",
  "codex": "CRITICAL",
  "antigravity": "BLOCKED"
}

GET /v1/providers/:name

Full ProviderMetrics object for a single provider. Returns 404 if unknown.

GET /v1/models/best-match

Returns the two coding-assistant models with the most depleted quota — useful for automatically routing tasks to the model with the most headroom. Excludes non-coding providers and informational-only metrics.

{
  "first": {
    "provider": "claude",
    "modelId": "claude-5h-window",
    "modelName": "Claude Session",
    "remaining": 44,
    "total": 100,
    "type": "percent",
    "health": "OK",
    "resetAt": "2026-06-01T13:40:00Z"
  },
  "second": { "..." }
}

GET /health

Liveness check — returns { "status": "ok" }.

---

Setup

git clone https://github.com/simonecamerano/ai-control-tower.git
cd ai-control-tower
npm install
cp .env.example .env
# Fill in credentials for the providers you use

Running

Command	Description
./launch.sh	Start server and open dashboard in browser
npm run dev	Start server in watch mode only
npm run build	Compile TypeScript to dist/
node dist/index.js	Run compiled production build

Server listens on http://localhost:3000 by default.

Testing

npm test             # run suite once
npm run test:watch   # watch mode

---

Adding a New Connector

// 1. Create src/connectors/myprovider.ts
export class MyProviderConnector extends BaseConnector {
  constructor() { super('myprovider'); }

  protected async fetchMetricsRaw(): Promise<ProviderMetrics> {
    // fetch quota data, return ProviderMetrics
  }
}

Then register in src/connectors/index.ts, add env vars to src/config.ts and .env.example, and write src/connectors/myprovider.test.ts. The dashboard and API pick up the new provider automatically on next restart.

---

Project Structure

ai-control-tower/
├── src/
│   ├── index.ts                    Fastify server entry point
│   ├── config.ts                   Environment variable loading
│   └── connectors/
│       ├── base.ts                 Abstract base + TTL cache
│       ├── factory.ts              Static connector registry
│       ├── index.ts                Connector registration
│       ├── claude.ts
│       ├── copilot.ts
│       ├── deepseek.ts
│       ├── tavily.ts
│       ├── codex.ts
│       └── antigravity.ts
├── public/
│   ├── index.html                  Dashboard SPA
│   ├── app.js                      Polling + render logic
│   └── style.css
├── docs/
│   └── screenshot.png
├── .contextforge/                  Project memory layer (ContextForge)
├── .env.example
├── launch.sh
└── roadmap.md

---

Disclaimer

This project is a personal tool built for educational and demonstration purposes. It performs no write operations, purchases, or actions on any provider platform — only reads quota and usage data that the authenticated user is already entitled to see through their own accounts. If you use this project, you are responsible for ensuring your use complies with the terms of service of each provider you connect to.

---

Author

Simone Camerano — AI workflow engineer and full stack developer.

• 🌐 simonecamerano.dev
• 💼 linkedin.com/in/simone-camerano
• 🐙 github.com/simonecamerano

---

License

MIT