The agent accountability layer.
AI Agent Audits.
Structured. Traceable. Evidenced.
Self-serve sign-up is open now. No invite code, no waitlist.
See it in action
Jiminy, in motion.
Not features, guarantees
Independence, Attestation, Reliability.
Three architectural guarantees that together close every known attack surface on AI accountability evidence.
Independence
The evaluator can never be the agent owner, enforced as a hard schema validator, not a policy or a checkbox. Closes the self-certification attack surface.
Attestation
Every step is signed into a tamper-evident chain. Any mutation after submission is instantly detectable. Closes the retroactive-modification attack surface.
Reliability
Every audit runs multiple independent judge passes; the most cautious result prevails. Closes the run-selection-gaming attack surface.
Live audit
Every trace runs through six criteria
EVALUATE
Every trace assessed against
accountability criteria in seconds
accountability criteria in seconds
ATTEST
Cryptographic chain of evidence
Tamper-evident and permanent
Tamper-evident and permanent
REPORT
Regulator-ready PDFs generated
for every decision on demand.
for every decision on demand.
The problem
You're deploying AI that makes consequential decisions. Can you explain every one of them?
The EU AI Act, FCA Consumer Duty, and ICO guidance all assume you can. Most teams today cannot. Jiminy closes that gap, without changing how your agents work.
SDK-based integration
Build a signed trace of your agent's decision with the Jiminy Python SDK and submit it for audit. No changes to your agent architecture required.
Criteria-based audit
Every trace is scored against a structured six-criterion rubric: scope adherence, tool authorisation, escalation judgement, output traceability, data boundary, and differential treatment.
Attestation chain
Signed step hashes link every tool call into a tamper-evident chain. Mutation is instantly detectable.
Multi-run reliability
Opt in to up to 5 independent judge passes on a single trace. Jiminy returns the majority verdict, with ties always resolved toward the more cautious outcome, by design.
In action
The audit report your compliance team has been asking for
Full dashboard: audit report, verdict banner, attestation chain, criteria breakdown
Works with your stack
Go deeper
Read more before you decide.
How it works
The actual architecture: agent decides, trace submitted, audit runs, report issued. Under 60 seconds.
EU PLD
9 December 2026, the rebuttable presumption applies. What that means for evidence, and why a dashboard isn't it.
Independence
Why an evaluator inside your own environment can't claim independence, and what Jiminy does differently.
Get started
Built for the people who hold AI decisions accountable.
Compliance leads. Legal teams. Internal audit. The FCA. Anyone who needs to look a decision in the eye and explain it.