Think. Plan. Stay in control.
Verified AI knowledge, scoped local Guardian tool permits, saved task plans, and text chat. External agent actions remain disabled.
Company-aware chat: ask naturally about NorthViax wholesale business, NrtVix AI, their relationship or official links. Answers come from a reviewed public brief, not invented by the model.
Intelligence that asks before it acts.
Describe a task. NrtVix proposes one safe local read; you review the exact scope, then Guardian verifies a fresh one-time permit before execution. No external tools, no autonomous loop.
Guardian review · no action yet
Execution & audit receipt
No execution. Existing accounts and tools remain unchanged.
Local M6 decision history
Security boundary: the model never receives API tools or access to the M5 executor. The server may authorize only public product facts and your own authenticated workspace summary. This is a local experimental integration, not production or the v1.6 external tool gateway.
Guarded local execution
This M5 tool flow uses a one-time, server-issued permit for scoped local reads. Every M5 decision is recorded. No external tools are connected.
Decision & audited receipt
Choose a read-only tool, then request a permit.
Local security audit
Text replies only. External actions are disabled.
From chat to a controlled plan.
Build a structured task plan, review the exact steps and Guardian decisions, and save a draft. Plans NEVER execute automatically.
Plan & Guardian decisions
Not generatedNo task plan yet. Start with a goal above.
Saved plans
Your drafts are stored in a separate local SQLite database and scoped to your logged-in workspace.
Advanced · Audited public facts lookup (optional)
AUDITED LOCAL TOOLVerified NrtVix product facts
Only the read_nrtvix_public_facts tool is allowed. No web browsing, email, payment, shell, database changes or file reads. Every attempt is logged.
Local Guardian audit trail
Checking configuration…
Offline preview; nothing is sent to OpenAI.
No tools are exposed to the model. M5 local read-only tools require a separate server-side permit; no external actions.