Monitoring Your AI Usage Through MonoRouter's Dashboard

When you route AI requests through MonoRouter, you get visibility into how your team actually uses Claude. The dashboard shows what matters without overwhelming you with data.

Request Volume

The main dashboard view shows requests over time — how many, which model, and whether they succeeded. Here's what a typical week looks like for a 4-person team:

Day	Sonnet	Opus	Haiku	Total	Success Rate
Mon	342	45	18	405	99.5%
Tue	298	52	22	372	99.2%
Wed	410	61	15	486	98.8%
Thu	387	38	20	445	99.7%
Fri	265	29	12	306	99.3%
Sat	42	8	0	50	100%
Sun	15	3	0	18	100%

Spikes tell you when your team is busiest. Dips might mean a session dropped out of the pool. A sudden drop in success rate is a signal to check session health.

You can filter by model (Opus, Sonnet, Haiku) to understand which models your team relies on most. This is useful for deciding whether Pro or Max subscriptions make more sense.

Token Consumption

Every response includes token counts. MonoRouter aggregates these into daily and weekly summaries:

Metric	Daily Avg	Weekly Total
Input tokens (Sonnet)	1.2M	8.4M
Output tokens (Sonnet)	180K	1.26M
Input tokens (Opus)	520K	3.64M
Output tokens (Opus)	95K	665K
Equivalent API cost	$48.20	$337.40

The "equivalent API cost" row shows what you'd be paying if these requests went through the Anthropic API. It's a running reminder of what MonoRouter saves you.

High input-to-output ratios (>5:1) usually mean your tools are sending large codebases as context. This is normal for Claude Code and Cursor, where entire files get included for context.

Session Health

The session pool view shows each connected account's status:

┌─────────────────────────────────────────────────────┐
│ Session Pool                                         │
├──────────────┬──────────┬────────┬──────────────────┤
│ Account      │ Status   │ Active │ Last Request     │
├──────────────┼──────────┼────────┼──────────────────┤
│ dev-alice    │ ● healthy│ 2      │ 3 seconds ago    │
│ dev-bob      │ ● healthy│ 0      │ 1 minute ago     │
│ dev-carol    │ ◐ throttl│ 0      │ 45 seconds ago   │
│ ci-pipeline  │ ○ disconn│ 0      │ 2 hours ago      │
└──────────────┴──────────┴────────┴──────────────────┘

• Healthy sessions are actively accepting requests
• Throttled sessions have hit a rate limit and are cooling down (auto-recovers)
• Disconnected sessions need attention — usually a browser session that expired

API Key Activity

If you've issued multiple API keys (one per developer, one per CI pipeline), the dashboard shows activity per key:

Key Name	Requests (24h)	Tokens (24h)	Last Active
alice-dev	186	2.1M	2 min ago
bob-dev	142	1.4M	15 min ago
ci-review	87	980K	1 hour ago
staging-tests	23	210K	6 hours ago

This helps identify which workflows are heaviest users and whether any keys might be compromised — an unexpected spike from a key that should be idle is worth investigating.

Setting Up Alerts

MonoRouter can notify you when things need attention. Configure alert rules in your account settings:

Alert Rules:
├── All sessions disconnected    → Email + Slack webhook
├── Success rate < 95% (5 min)   → Slack webhook
├── Single key > 500 req/hour    → Email (anomaly detection)
└── Session throttled > 10 min   → Dashboard notification

Each alert supports multiple notification channels. Most teams set up a Slack webhook for real-time alerts and email for daily summaries.

What We Don't Track

Privacy matters. MonoRouter tracks operational metadata — timestamps, token counts, model names, status codes — but never the content of your prompts or Claude's responses. Your code, your conversations, and your intellectual property stay between your SDK and Claude.

The dashboard shows you how much you're using Claude, not what you're using it for.