DADI-OPS v0.1.0 · Prototype and pilot delivery; interface data is illustrative Delivery boundary & compliance →
Capabilities

What the operations layer does

DADI-OPS is not another video wall. It turns day-to-day security operations into executable, attributable actions. Each capability below states what it solves and what it deliberately does not do.

11 views6-step loop3 rolesSHA-256 auditlocal-first AI
01 / SITUATION

See the state before acting

Device scale, uptime, alarms and tickets on one screen, so a shift handover does not require five spreadsheets.

Operations cockpit

Six KPIs plus a 7-day alarm trend, zone uptime, live alarm stream and state composition. Jump from any metric straight to the device, alarm or ticket list.

Metric basis: device totals and uptime come from the access layer; alarms and open tickets are counted live by the operations layer.

Analytics wall

Progress rings, uptime and inspection trends, alarm and source distributions, ticket funnel, quality tiers and vendor distribution. All rendered as local SVG, with no CDN or build chain.

No invented history: trends only come from real inspection runs; empty data shows a placeholder, not a fabricated curve.
02 / SIGNAL

From alarm flood to a workable event

The problem is rarely too few alarms; it is the same fault reported dozens of times.

Clustering and correlation

Same-device, same-type alarms are clustered within a 30-minute sliding window. Events retain every original alarm ID. Multi-device outages in one zone suggest an upstream or power fault; short high-frequency bursts from one device indicate repeated reporting.

Boundary: clustering only covers duplicate reporting. The compression rate is computed live and may be zero before real alarm traffic arrives.

Advice and repair wizard

Steps, required parts, estimated effort, priority, evidence and confidence. The wizard classifies an event as remote-only, on-site, remote-then-site, or insufficient evidence, and returns structured steps with expected observations.

Boundary: no remote commands are executed and nothing is dispatched automatically. Repair records enter the audit chain.
03 / LOOP

Make the action an acceptable ticket

Alarm-to-ticket is not copy-paste; it is a business object with a state machine, an owner and a write-back relationship.

Ticket board

Pending dispatch, in progress, pending acceptance and closed, with priority, owner and SLA. Closing a ticket archives the linked alarm and records the actor and timestamp in the audit log. Repeated conversion of the same alarm is idempotent.

Inspection and duty

Plans and runs are separate; a full inspection produces a real uptime time series and turns anomalies into alarms. Built-in three-shift duty rosters annotate alarm closure, ticket creation and inspection anomalies. Notifications stay in local logs unless a webhook is explicitly enabled.

Security: production should add an internal-address allowlist to the webhook to prevent SSRF.
04 / ASSETS

Know what you have, and whether it still works

The registry answers “what is out there”; quality checks answer “is the picture usable”; the map answers “where is it”.

Device monitoring and quality

Device status, uptime, heartbeat and channel count with search and filters. Five quality findings — black screen, blur, noise, freeze and signal loss — with rule-based review against device state and alarms.

No overclaiming: without an image model, the system does not claim to have recognised the picture.

Asset registry and map

Model, vendor, location, warranty and service provider in one registry, reused by risk ranking, vendor reports and map markers. The schematic map works fully offline; the satellite base map is optional and falls back automatically.

Compliance: no map key is shipped in the front end; demo coordinates are explicitly marked.
05 / TRUST

Every write can be questioned

An operations system must not only act, but prove who did what and when.

Authentication and RBAC

Password login with scrypt and random salt, hashes held in memory only; session tokens from a cryptographic RNG with an 8-hour TTL and immediate logout revocation. Five failures lock the account for one minute; unknown accounts and wrong passwords return the same message.

Audit hash chain

Each audit entry carries a sequence number, previous hash and its own SHA-256. Verification recalculates the chain and locates field edits, middle deletions, tail truncation, reordering and forged sequence numbers. An anchor parameter detects whole-file rewrites.

Key boundary: a local chain detects tampering but cannot stop someone who can rewrite the file and recompute the chain. Production must anchor the daily head to independent read-only storage and retain six months.
About the word “AI”: DADI-OPS uses a rule engine with explainable evidence, not a chat mascot. It is local by default with zero external requests. An optional LLM only improves wording on redacted text. All write operations require human confirmation; the system never auto-dispatches or auto-closes.

Want to walk the real chain at your scale?

Share device count, vendors and current access method; we will scope a pilot and provide an acceptance checklist.