Promptfoo describes itself as “AI security testing: "Ship agents, not vulnerabilities" - automated testing that finds and fixes AI risk in development”. COLVO answers a narrower question: did your agent’s refund or cancellation actually happen correctly in Stripe — tested before release, guarded and verified live.
| COLVO | Promptfoo | |
|---|---|---|
| What it is | Testing and live guarding for AI agents that refund, cancel or change customer accounts — judged by the provider state, not the reply. | AI security testing: "Ship agents, not vulnerabilities" - automated testing that finds and fixes AI risk in development. source ↗ |
| Main features |
|
|
| Checks the payment provider’s state after an agent action | Yes — every action is re-read and marked verified, mismatch or unverifiable | Not described on their public pages |
| Holds a live action for human approval | Yes — above a threshold set in the mandate, before anything reaches Stripe | Not described on their public pages |
| Open source | Hosted service; SDKs on npm and PyPI | Partly — open-source libraries plus a commercial platform |
| Pricing | Free · Test €19 · Guard €99 per month (pricing) | Community edition free (up to 10k probes/month); Enterprise and On-Premise are custom pricing. source ↗ |
| Best for | Teams whose agents move money or change subscriptions, and who need proof it went right | Developers and security teams who want to red-team and evaluate LLM apps and agents for vulnerabilities before shipping. |
Facts about Promptfoo come only from their own public pages (linked above), last verified on Sep 30, 2026. “Not described on their public pages” means we did not find it there — not that it is impossible. Promptfoo is a trademark of Promptfoo, Inc.. Spotted something out of date? Tell us and we will fix it.
Keep Promptfoo for what it is built for. Add COLVO where a wrong action costs money: it tests refunds and cancellations in a sandbox that fails on purpose, then guards and verifies the live calls.