Ziwei Guo — Founder, Witnora
Safeguards for AI agents handling refunds and order cancellations.
I build Witnora to help teams put clear limits around AI actions that affect real customers. We check agreed permissions before an action runs, control execution through a customer-owned boundary, and independently check the result in the business system. A signed receipt connects the permission, execution, and observed result.
See a refund walkthrough · Explore the open-source framework · Contact me
The walkthrough is a recorded test-mode replay of retained evidence; it issues no new refunds.
I'm looking for early design partners to test one refund or cancellation workflow at a time. I build the product and work directly on integrations and technical support.
For engineers: verify the evidence yourself
The Python verifier and test vectors are open under Apache-2.0. After obtaining the source, verification runs locally with no Witnora account or network connection.
git clone --depth 1 https://github.com/witnora/witnora-framework
cd witnora-framework/packages/witnora-verifier-python
python -m unittest discover -s tests
PYTHONPATH=src python -m witnora_verifier.cli conformance ../../test-vectors/witnora-v0.2-conformance- Offline verifier (Python) — canonical JSON, Ed25519 signatures, and conformance tests.
- Trust bootstrap — pin trusted roots separately; an evidence packet cannot authorize its own root.
Signature verification checks record integrity and authenticity relative to the trusted key. Outcome verification depends on the separate observation and its configured scope.
Coverage applies only to connected, explicitly configured action paths. Witnora does not guarantee all future agent behavior or claim third-party certification.