Live — test your agent now

Does your AI agent handle money safely?

Canary runs 5 financial attack scenarios against your agent and returns a trust score in under 30 seconds. No SDK. No setup. Just paste your system prompt.

Test Your Agent →
How It Works

Three steps to trust

Paste your agent config, we simulate real financial attacks, you get a reliability scorecard.

1

Paste Your Prompt

Drop in the system prompt and model your AI agent uses. That's all we need.

2

We Attack It

5 predefined financial scenarios test overspend, duplicates, compliance, rate abuse, and error handling.

3

Get Your Score

Pass/fail per scenario plus an overall trust score 0–100. Know exactly where your agent breaks.

Scenarios

What we test

Five scenarios that catch the failures that matter in production.

💸

Overspend Protection

Can your agent refuse a transfer that exceeds the account balance?

Safety
🔀

Duplicate Detection

Will it catch the same purchase request sent twice in 30 seconds?

Reliability
🚫

Unauthorized Vendor

Does it block payments to vendors on the company compliance blocklist?

Compliance

Rate Limit Abuse

Can it flag 8 rapid-fire transfers that deviate from normal patterns?

Safety

Timeout Resilience

When a payment times out, does it blindly retry or check status first?

Reliability
Test Your Agent

Run the Canary suite

Paste your agent's system prompt. We'll run all 5 financial scenarios and return a scorecard.

The instructions your AI agent operates with.
Running 5 financial scenarios...
Overspend Protection
Duplicate Transaction Detection
Unauthorized Vendor Block
Rate Limit & Rapid-Fire Detection
Timeout & Error Resilience
Trust Score
Passed
Failed
Duration

Ship agents you can trust.

Canary is free during early access. Test as many agent configs as you want.

Test Another Agent →