Trelo is in BETA. One line of code to stop agent loops, duplicates, and attacks. Join early access
Stop infinite loops, prevent attacks, and cut token waste by 40-70%. One line of code.
Trust middleware
for AI agents
Eight problems.
One proxy.
Infinite Loops
Agents hammer the same broken tool call hundreds of times while you watch the meter run.
Prompt Injection
Malicious user input tricks your agent into deleting data or calling unauthorized APIs.
Runaway Spend
You discover the $4,000 API bill on Monday morning with zero idea which agent caused it.
Duplicate Actions
The same agent charges a customer twice because it lost track of what it just did.
No Side effects safety
Emails send twice. Orders place twice. Database writes happen twice. Every retry is a disaster.
Black Box Audits
An agent makes a bad decision at 3 AM and you have zero logs to explain why.
Silent Failures
The agent stops responding mid-task. No error. No timeout. Just silence and a stuck customer.
Retry Storms
One failing API triggers a cascade of retries that burns your monthly budget in 20 minutes.
BY THE NUMBERS
What we're seeing in private beta.
Two steps.
Zero config.
Install the guard
Setup onceOne command. Python, FastAPI, SQLite. No external services, no config files, no infrastructure changes.
pip install treloWrap your agent
RuntimeAdd one line to your agent code. Circuit breakers, dedup, firewall, and audit trail activate immediately.
agent = Agent(guard="trelo")Start free.
Scale when ready.
Free
$0
Pro
$29/mo
Team
$299/mo
Open source? Self-host Trelo for free. MIT licensed. GitHub →
One line of code.
Eight layers of protection.
Your agents should ship fast, not burn your API budget at 3 AM.