Give humans the final say on risky actions.
GateLedger is a local policy authoring and simulation workbench. It cannot intercept, authorize or block actions in a running agent. Once you export rules, your engineering team must enforce them in a trusted system.
If there is no active matching policy—or a permission has expired—the simulator returns BLOCK by default. This is a planning convention, not a guarantee of security in a production agent.
1. Inventory every tool action
Record the agent/tool, whether it reads or writes data, what it touches and whether the effect can be reversed. This is more useful than giving one broad permission to the entire agent.
2. Choose the decision for each action
- Allow: appropriately scoped, low-impact actions where automatic execution is acceptable.
- Require review: sending messages, changing records, publishing, spending or making irreversible changes when an accountable human must approve.
- Block: actions outside approved scope or unacceptable risk.
3. Assign ownership and review expiry
Add an accountable reviewer and a review-by date. Expired entries cannot match in this simulator. A real integration should handle authentication, approval deadlines, revocation and authoritative logging.
4. Run hypothetical tests
Use the simulation panel to see the planned decision for an action. Export JSON for engineering and CSV for a reviewer; the exported file contains definitions, not enforcement code.
5. Implement enforcement outside this website
A trusted workflow runtime must check the current permission and user identity before executing a tool call, wait for genuine reviewer authentication, apply the decision, handle timeouts/retries and save an audit trail. Protect access to the policy file and its rollout process.
Research and background
- Production n8n human-review discussion (August 2026)
- Human approval workflow example (November 2026)
- Example of a live approval platform with API/MCP integrations
Sources inspected November 16, 2026. Community examples are not independently audited implementations.
Common questions
Does GateLedger block a real AI agent action?
No. It is a local planning and simulation workspace. Production approval enforcement requires a trusted integration you build and test separately.
Can I audit an n8n or MCP workflow automatically?
No. This edition does not parse live agent configurations or connect to those services. You enter policy rules manually.
What happens after a permission expires?
The local simulator returns BLOCK for expired rules until an active matching rule exists. This is a planning convention and does not affect real workflows.
Where are my policy rules saved?
They are stored in this browser's localStorage, not in a cloud account. Export JSON for backup or handoff.
Design a policy ↗