NeuroVaultAGENT GOVERNANCE · POLICIES
Understand what your assistants can do.
These rules decide whether requests sent through NeuroVault are blocked, need approval, require masking, or may proceed. An Allow rule alone does not grant an assistant access to a tool.
Checking your session…
Current permission rules
Previous versions
Choose a protection preset
Prepare one rule for one assistant. Existing rules stay in the draft, and nothing changes until you review and publish.
Matching requests will be blocked. Requests with no matching rule remain blocked by default.
Invite a real colleague with reviewer access before using this rule. You cannot approve your own assistant. Reviewer invitations are not available yet; an administrator must arrange access. If you work alone, choose Block or Hide sensitive data, or try the approval demo.
Try the approval demoAdministrator tools: edit and publish permission rules
Manage organization policy
Edits below are an unpublished draft. Publishing replaces the active rule set and retains all previous versions. Priority is Block, Require approval, Redact, then Allow. No matching rule means Block.
Approval on a tool response withholds that response; response approval and release are not implemented. See the Agents page for configured tool names and request destinations. The response destination is agent. No external model is connected.
Optional: simulate a policy decision without executing a tool
Policy simulation
Use an agent you registered. Category labels are hypothetical, not detector results. Exact labels and capitalization matter.