I built a smoke detector for AI agent wallets
I built a spend monitor for AI agents making autonomous payments AI agents on x402/stablecoin rails can make hundreds of payments a minute with no human approval. There's no equivalent of a bank fraud alert for this yet
I built a spend monitor for AI agents making autonomous payments
AI agents on x402/stablecoin rails can make hundreds of payments a minute with no human approval. There's no equivalent of a bank fraud alert for this yet โ if your agent goes rogue or gets compromised, you find out when you check your balance.
Burnwatch is a thin Python SDK shim that mirrors payment metadata to a backend that learns each agent's normal behavior and alerts when something looks wrong.
What it detects: velocity spikes, drain bursts, unknown payees, new rails, off-hours spend, daily budget overruns, blocklist matches โ 13 rules total, all transparent with human-readable evidence.
Key constraint I kept: observe-only. It never touches keys or sits in the payment path. Fail-open โ if my backend goes down, your agent keeps paying. I wanted this to be a smoke detector, not something that could accidentally block legitimate payments.
Stack: FastAPI + PostgreSQL + React, hosted on Linode. Python SDK on PyPI.
pip install burnwatch
Launched today. Free during beta. Would love feedback โ especially from anyone building on x402 or running agents that make real payments.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes โ full credit and traffic to the original publisher.