A local sentinel for autonomous AI: high-risk actions wait for your approval, secrets are auto-redacted, and every action is written to a tamper-proof hash-chained audit log.
Make Autonomous AI Safe to Trust, Audit and Roll Back
Core Capabilities
High-Risk Actions Queue Up — AI Waits for Your Nod
When AI tries to spend, order or send outbound, it queues for your approval — no money moves without your explicit nod
Approve or Reject in One Line
Send 'Pending' to see the queue, send 'Approve 2' or 'Reject 2' to act — manage all AI autonomous actions from WhatsApp, no dashboard needed
API Keys & Tokens Auto-Redacted — Never Logged in Clear Text
Regardless of whether AI actions carry credentials, Warden auto-detects sk-xxx, Bearer xxxxx and similar formats and redacts them — no full key ever appears in the audit log
SHA-256 Hash Chain: Tampering Is Detectable
Every action is hashed against the previous entry, chained together — any record modification breaks the chain, instantly flagged when you run 'Audit'
Today's Action Summary + Chain Integrity in One Check
Send 'Audit' to see today's action count, approval/rejection tally and chain integrity in one shot — a 30-second daily review of all AI activity
Runs Entirely on Your Box — Audit Data Stays Local
Both the approval queue and audit ledger are stored locally on your Zhimao Box — no third-party cloud; even offline, the full history is intact and readable
Roadmap
Requirements
- ✓Zhimao account with AI enabled
- ✓Zhimao Box (approval queue and audit ledger stored locally)
- ✓Best used with spend Mandates (AP2 authorization) for a full trust loop
Updated 6/14/26, 11:12 AM