AI acceptance testing
Ship gate for AI
Let the monkeys stress it before your users do.
Your agent shipped a feature. Did it actually do what you told it to? Crazy Monkeys returns a versioned acceptance verdict with evidence: ship or don’t, before merge.
Built for teams and solo builders shipping AI apps, agents, and tool-calling features.
Proof you can trust
- Baseline on your preview or PR
- Catch tool-call loops and injection nonsense
- Ship / no-ship keyed to the build (SHA)
- Evidence you can keep: not a script dump
What you get
One run answers, for a specific workflow on a specific commit:
- What it was supposed to do
- What “correct” looks like
- Evidence (telemetry / traces / captures)
- Pass or fail
- Diff since the last run on the same step
- SHA the verdict was produced against
The runner underneath can change. The contract shape does not. The script is interchangeable. The ledger is the product.
What we are not
- Not QA automation theater
- Not an AI bot that authors your Playwright suite as the SKU
- Not a coverage-as-a-Service staffing shop
- Not enterprise SSO / on-prem (stubs stay stubs)
How it fits CI
Point us at a preview. We run acceptance. You get ship or don’t ship before merge. Limited beta. Waitlist open.