MIMI is the aggregation & correlation layer for NOC teams — reading from NetBox, Prometheus and Icinga to turn a storm of signals into one incident per root cause, with AI-assisted investigations and a governed action gateway on top.
Every device, metric and check flows into one model. MIMI writes only to its own audit log — your inventory and monitoring stay the source of truth.
A local model walks the blast radius through a tool-calling loop and returns ranked causes with citations. Persistent, auditable, escalatable to a frontier model.
Least-privilege probes identify a new device, then the model drafts a reviewable onboarding diff — observed committed as-is, inferred flagged for edit.
Versioned config capture with running-vs-baseline drift detection, per-vendor normalization, and diffs pinned to the incident that touched them.
One mutation path recomputes each action's tier from blast-radius × reversibility. T1 confirms, T2 needs an admin, T3 needs two plus a ticket.
A live pulse band — fleet health, severity mix, activity — over a real geo map of your sites. Every widget is role-aware and rearrangeable.
Resolutions promote into runbooks and known-issues that ground the model (RAG) and match new incidents against what you've already solved.
OIDC sign-in with Google Workspace and Microsoft Azure AD, coexisting with local accounts.
Self-registration is off by default. Accounts are provisioned by an administrator or your IdP.
The first account to register becomes the initial admin — permanently immune from modification by other admins.
Every mutation is tier-checked at one gateway; the proposer can't self-downgrade its risk.
High-blast changes need two distinct admins, a linked ticket, and respect per-site change windows.
Every attempt — including denials — is logged with actor, approver(s) and requested-vs-enforced tier.
Sign in with your organization identity, or set up the first administrator to get started.