agentsitter 0.1.0

The babysitter for your AI agents. Watches costs 24/7 and alerts you if something goes wrong.

Agentsitter 🟢

The babysitter for your AI agents.

Agentsitter watches your AI agents while you sleep. If something goes wrong — a loop, a runaway cost, anything odd — you get alerted before it costs you thousands.

Install

pip install agentsitter

Usage

Add three lines to your existing agent:

import agentsitter

sitter = agentsitter.watch(
    api_key="agt_your_key",   # Get your key at agentsitter.net/start
    alert="[e-mail]"
)

# After every API call, track the cost:
sitter.track(cost=0.04)

That's it. Agentsitter runs silently in the background, checks your agent's behaviour every 60 seconds, and alerts you if anything looks wrong.

What it watches for

• Loops — your agent repeating the same call over and over
• Runaway costs — spend accelerating exponentially
• Statistical drift — behaviour that looks abnormal based on Lo-MacKinlay Variance Ratio analysis

Get your API key

Go to agentsitter.net/start to get your free API key. Takes 30 seconds.

Daily digest

Once set up, you'll get a daily email at 8am with your agent's health status:

🟢 Status: GREEN
Your agent looked healthy today.
API calls: 847 | Spent: £2.41 | Anomalies: None

Links

• Website: agentsitter.net
• Live demo: arena.trustlogdynamics.com
• GitHub: github.com/AnouarTrust/agentsitter

Built in Manchester

By Anouar Boussif · Comptex Labs