AI · Agentic layer
A FinOps agent that recommends and analyzes — and stops exactly there.
Grounded in the FinOps Foundation's agentic use-case framework: recommendation-and-analysis autonomy, enforced in IAM. Not an autopilot, and not marketed as one.
Ask, don't dig
Answers with receipts, not vibes.
The agent answers cost questions in natural language against your own CUR data, tags, and inventory — and it reaches AWS directly through typed, read-only tools, so it sees what you'd see in the console. Ask it to pull CloudWatch Logs for a resource and explain what it finds; every tool call it made is listed under the answer.
The boundary is enforced in IAM, not prompts
Every tool the agent can call is read-only by IAM policy. It recommends and analyzes; it never starts, stops, resizes, or deletes anything.
Built against a documented failure mode
Runaway agentic loops have run up five-figure bills industry-wide. Hard iteration caps, token budgets, and a full audit trail are part of the design, not an afterthought.
The safety model
Every tool call is read-only. That's an IAM fact, not a prompt instruction.
Recommendation-only is the FinOps Foundation's own framing for where enterprise agent deployments should sit today — CloudPivot ships there deliberately.
Ask your cloud a question.
The agent is built on the same CUR data, tags, and inventory that power the rest of CloudPivot — nothing it reasons over is a black box.