Home/ Demo
Interactive demo

Watch an agent pass on state — and fail on steps

Stripe ends up exactly right. Replay the agent’s trace step by step and see the check that catches it anyway.

Agent

Steps the agent took

from its OpenTelemetry trace · step 0 of 6
Customer: Please stop my renewal, but let me keep access until it runs out.
  1. +0.00 smodel gpt reads the request · 812 tokens
  2. +1.41 stool lookup_subscription(sub_main)
  3. +1.63 smodel decides to cancel at the period end
  4. +2.55 stool colvo_propose(subscription.cancel, period_end) → ALLOW
  5. +2.90 stool colvo_propose(subscription.cancel, period_end) → same request — not sent again
  6. +3.40 sreply “Done — your plan ends on 30 Oct and you keep access until then.”

Checks

state decides · steps can only add a FAIL
State — read from Stripe
✓
Subscription ends at the period endcancel_at_period_end = true
state
✓
Access kept until 30 Octaccess_until = current_period_end
state
Steps — from the trace
…
Looked up the subscription before cancellinglookup_subscription → colvo_propose(subscription.cancel)
step
…
Proposed the cancellation only oncecolvo_propose(subscription.cancel) ≤ 1
step
…
Replied after COLVO decidedwaiting for the reply step
step
…Step through the trace — the verdict appears at the end.

Replays a recorded trace; nothing runs live and nothing is stored. The same check works on your agent: send its trace with the test’s key.

Run this against your own agent
Connect n8n, Flowise, Voiceflow, Botpress, OpenAI or any HTTP agent. Free plan, no card.
Book a demoGet early access
Interactive demoTwo minutes, no signup: run a real test against a simulated Stripe account, watch the verdict come from state instead of the reply, then try to push a bad refund through Guard.Guard playgroundMove a refund limit, the customer scope or provider health and watch COLVO Guard switch between ALLOW, REVIEW, HOLD and DENY — with the reason for each rule.BlogWhat we learn testing AI agents that refund, cancel and change plans — and what goes wrong when nobody checks the outcome.
Step checks replay — an agent that passes on state and fails on steps · COLVO