Runtime AI agent monitoring

Your AI just made a request.
Do you know what was in it?

IcyHat is a dashcam for your AI. See every network request, token count, and dollar spent — across Claude, ChatGPT, Gemini, and more. Monitor, alert, or block in real time.

IcyHat — claude.ai
14:32SAFEChat message via sonnet-4-5 — 2.1K tokens — $0.008
14:33ALERTContext surge — history jumped +18K tokens in one turn
14:35SAFETool use: web_search — 4.8K tokens — $0.02
14:41BLOCKRapid-fire loop — 6 requests in 8s — Paused for review

AI agents are doing more than you think

Every message you send triggers network requests you never see. Token counts climb silently. Agents use tools, switch models, and accumulate context — all invisible.

Power users run 5, 10, even 15 agents simultaneously with zero monitoring. Some blow past $200 in API costs in two days without noticing.

0
tools monitoring
consumer AI agents
100%
of agent requests
are invisible to you
💸

Cost blindness

Nobody knows what their AI usage actually costs until the bill arrives. Platforms obscure thinking tokens and model switches.

🔄

Silent spirals

Agents loop, retry, and snowball context without warning. A simple question can balloon into 100K+ tokens.

🤖

Autonomous actions

Agents search the web, run code, access files, and call APIs — through requests you never see or approve.

📊

Dead context bloat

Stale tool results from early turns get resent every message. You pay for the same data over and over.

Transparent layer. Zero interference.

IcyHat sits between your AI interface and the network. Normal chat flows untouched. Suspicious activity gets flagged — or frozen.

01

Install and forget

Add IcyHat from the Chrome Web Store. It activates automatically on supported AI platforms. No configuration needed.

02

See everything

Every request is logged with model, token count, estimated cost, and behavioral analysis. A slim panel shows activity in real time.

03

Take control

Smart thresholds flag unusual activity automatically. Large requests, rapid-fire loops, and context surges get caught. You decide: allow or block.

Your monitoring, your way

Choose the level of control that fits your workflow. Switch anytime.

🟢

Monitor

Smart thresholds handle everything. Normal chat passes silently, alerts fire on unusual patterns, suspicious activity gets blocked.

RECOMMENDED
📊

Token Meter

Pure visibility. Every request logged and displayed, nothing blocked. Awareness without intervention.

PASSIVE
🛑

Firewall

Full lockdown. Every outgoing request is paused until you explicitly approve it. Maximum control for sensitive work.

MAXIMUM CONTROL
Built for power users who run real workloads

Everything you need to understand and control what your AI agents are doing.

🔍 Behavioral Analysis

Tracks tool use, model switches, context surges, and sustained autonomous activity across requests. Detects patterns that token counting alone misses.

🌀 Spiral Detection

Four-signal scoring catches agents in loops: turn ratio, similar-length responses, repeated tool calls, and snowballing sizes.

💀 Dead Context Alerts

Flags stale tool results and large responses from early turns that get resent every message — bloat you're paying for repeatedly.

💰 Model Comparison

Every request card shows the cheapest alternative model and your potential savings. See what switching would save.

📋 Context Handoff

One-click structured summary from your intercepted messages. Copy to clipboard, paste into a fresh chat. Stop paying for dead context.

📸 Shareable Rap Sheet

Canvas-generated PNG with session stats — tokens, cost, requests, behavioral signals. Auto-downloads for sharing.

📈 Message Fuel Gauge

Visual daily usage meter based on your plan limits. Auto-detects your subscription tier — no manual setup.

Smart Verdicts

Every request gets a verdict: SAFE, ROUTINE, CAUTION, or WARNING. Based on tokens, cost, behavior, and conversation patterns.

Works everywhere your AI lives

Monitors all major AI chat interfaces and local agent endpoints.

Claude
ChatGPT
Gemini
Copilot
Perplexity
Poe
You.com
Local agents
Start free. Upgrade when the clicking gets old.

Free tier gives you full visibility and manual control. Pro removes the friction.

Free
$0
forever
  • Full activity visibility
  • Behavioral analysis
  • Manual Allow / Block
  • All verdicts and alerts
  • All 8 platforms
  • Custom auto-rules
  • Budget hard-stops
  • Audit trail export
Install Free
Pro Annual
$39
per year — $3.25/month
  • Everything in Free
  • Custom auto-approve rules
  • Session budget hard-stops
  • Alert intensity control
  • Audit trail export
  • Priority support
  • Founding member perks
  • All future Pro features
Get Pro

First 500 lifetime deals available. One-time payment, yours forever — including every future feature.

Your AI agent is working
right now. Are you watching?

Install IcyHat in 10 seconds. See your first intercepted request before your next message.

Add to Chrome — Free