AI that is held to account.
Every AI in Nexus Lume has its own guardrails: a cost cap, a kill switch, four-eyes approval or a person who decides.
Four guardrails, written in the code.
Not a policy on paper: every rule runs in the product, before the AI acts.
Four-eyes approval
With a health, finance or legal compliance pack, Autopilot never publishes on its own, not even in autonomous mode. For a Pix payment (Brazil’s instant payments) sent inside a conversation, above the threshold, a second person has to approve.
Kill switch
Switched off, Autopilot runs no cycle at all, and the block is recorded. TIZ recaps don’t even switch on without a compliant AI provider.
Cost cap
Autopilot stops at the monthly cap the company sets. Every Lume Agents conversation has its own budget and is blocked once it goes past it.
A person in the loop
In copilot mode, every Autopilot piece waits for approval in the Sandbox. In candidate screening, the AI explains and a person decides.
Marketing · Autopilot · Activity
Every cycle, with its cost and reason.
Every Autopilot cycle is recorded with its stage, cost and duration; what the kill switch or the cost cap stops shows as blocked.
The company sets the monthly cap and the mode: copilot, with approval for every piece, or autonomous, always with approval under the regulated packs.
A blocked cycle: the cost cap was reached.Grupo Vértice · fictional company · illustrative data
Every AI, with its own guardrails.
Where AI works in Nexus Lume, what it does and what holds it back.
Marketing Autopilot
Researches trends, then writes, scores and schedules the brand’s content.
The guardrail: Compliance packs for health, finance and legal enforce forbidden terms, a mandatory disclaimer and four-eyes approval.
Lume Agents
Answer customers with a supervisor and a specialist for each topic; the company can switch on a reviewer before anything is sent.
The guardrail: Five layers of guardrails, the cost of every model call on record and a cap per conversation.
HR Copilot
Explainable candidate screening and pay equity analysis by group.
The guardrail: Every use records its legal basis and its fairness check. A person makes the decision.
TIZ recaps
Sums up channels and meetings for whoever joins later.
The guardrail: Without a compliant AI provider, the feature stays off.
Compliance MCP
An MCP server lets AI agents query the company’s controls and evidence.
The guardrail: Each agent sees only what its role allows, and every call lands in the audit trail.
AI assistants over the data of every product, in Data & AI. They join this list once they are in production.
Not every answer needs a model.
In the CRM and in Treasury, the answer comes from a score or a rule, not a language model, and the reason stays in view.
Commercial · Today
Every priority, with its reason.
Today: the rep’s queue in priority order, with the reason for each item. It is a weighted score, not generative AI: the factors stay in view.
Today’s queue, with the reason for each item.Grupo Vértice · fictional company · illustrative data
Lead scoring
Lead scores from 0 to 100, with the factors in view, so the sales rep understands the score.
Win probability
Win probability and close-date drift per opportunity, with stage anchors the company adjusts.
Treasury Copilot
Answers by reading Treasury’s real engines, read-only, with the source of every answer. It runs on rules, with no language model.
AI governance
In Compliance, a register to classify and review every use of AI in the company.
Questions about AI
Does the AI publish on its own?
Only if the company chooses autonomous mode, and never under a regulated pack. In copilot mode, every Autopilot piece waits in the Sandbox until a person approves it; with a regulated pack, nothing goes live without human approval, not even in autonomous mode.
Can it be switched off?
Yes. Switched off, Autopilot runs no cycle at all, and the block stays on record. TIZ recaps switch on only with a compliant AI provider.
How much can the AI spend?
Whatever the company decides. Autopilot stops at the monthly cap, and every Lume Agents conversation has its own budget.
Does the CRM use generative AI to set priorities?
No. The Today queue ranks the rep’s work with a weighted, explainable score and shows the reason for each item.
Who decides on a hire?
A person. Explainable candidate screening, with a human decision, and every use records its legal basis and its fairness check.