Know what your agents did. Prove it.
See every decision your AI agents make, stop the risky ones before they happen, and show anyone exactly what happened.
Claims triage agent held · A$7,494.31
Live from the public demosynthetic customerslast 24 hours120 runs, 3 stopped, 5 held
Trace.
Stop.
Fix.
Prove.
Step 1Request
Customer #8842 disputes an A$250 charge and wants a refund.
Step 2Modelscripted
lookup_customer(
customer_id: 8842) Step 3Toollookup_customer143 ms
tier: gold email: [redacted:email]
Step 4Modelscripted
issue_refund(
customer_id: 8842, amount: 250) Step 5Policy gateblocked
issue_refund(
customer_id: 8842, amount: 250) Step 6Run ends
stopped by a rule
- 1Start
- 2Think
- 3Lookup
- 4Decide
- 5Blocked
- 6End
- RecordSigned
Each step is fingerprinted over the one before. Change one byte of step 4 and every step after it shows it.
A scripted sample run. Its fingerprints are real.
Stop
Stop it, or hold it for a person.
This request is real, from the demo agents.
Fix
Change one line.
One question, put to gpt-4o twice.
Customer #8842$250 disputedgold tierrefund?
Customer #8842 disputed a $250 charge and is requesting a refund. They are gold tier. Decide the action.
systemYou are a careful enterprise agent. Follow policy.
Added to the system prompt: Never refund a disputed charge over $100 yourself — call escalate_to_human first.
Without the lineissue_refund()Blocked1,178 ms
With itescalate_to_human()Allowed1,228 ms
Real gpt-4o calls, 5 Oct 2026, quoted verbatim, so amounts read “$”.
Prove
Check it yourself.
The sample refund's record, checked on this server. Change one number and watch it break.
Try to break it
- Start
- Think
- Lookup
- Decide
- Blocked
- End
Chain check
Not checked yet.
The record checks itself when it scrolls into view.
Fits where your agents already run.
- TypeScript
- Open
Telemetry - Python
- Vercel AI SDK
- LangChain
- OpenAI
- Anthropic
For teams where a wrong decision matters.
From A$49/mo, or self‑host the source. Pricing →