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

Security operations
that actually close the loop.

DADI-OPS adds the missing operations layer on top of video access: device uptime, alarm noise reduction, ticket closure, inspection, quality checks, asset registry and a verifiable audit trail. Video access comes from a mature upstream; the operations business layer is built in-house.

GB28181-2016 access Zero runtime npm dependencies On-premise deployment
DADI-OPS / COCKPIT PLATFORM RUNNING
Device uptime
97.4%
illustrative
Alarms today
24
3 major
Open tickets
9
SLA running
7-day alarm trendcritical / warning / info
Interface illustration · demo data, not real operations v0.1.0
11 views
Covering the full operations workflow
10 phases
P0–P9 delivery phases, each verifiable
0 runtime
npm runtime dependencies
3 roles
RBAC enforced server-side
Why an operations layer

Getting video in is not the same as keeping it useful.

GB28181 access solves “can we see it”. The long-term cost sits in “can we keep it running”: who notices an offline camera, who reduces alarm floods, who closes the fault, and who can prove it afterwards.

01 / UPTIME

Offline devices found by luck

Hundreds of cameras across a site; an offline device is often discovered only when someone needs the footage.

Heartbeat and uptime on the cockpit, anomalies raise alarms
02 / NOISE

One fault, dozens of alarms

Flapping devices and unstable links generate floods; the one alarm that matters gets buried.

30-minute clustering, original alarm IDs retained
03 / EVIDENCE

Work scattered across chats

Who dispatched, who fixed, when it was accepted — no single record when an audit arrives.

Ticket closure plus a SHA-256 audit hash chain
Capabilities

One platform, one operational chain.

Eleven views share the same business state. Every conclusion carries evidence; every write operation leaves a trace.

Operations cockpit

Situation

Device count, uptime, offline count, alarms, open tickets and inspection completion on one screen, with trend, zone uptime and live alarm stream.

KPIuptimealarm stream

Alarm noise reduction

Signal

Same-device, same-type alarms are clustered into events within a 30-minute window; multi-device outages in one zone suggest an upstream fault.

clusteringinsighttraceable

Ticket closure

Loop

One-click conversion from alarm to ticket; four states with owner and SLA; closing a ticket archives the linked alarms.

dispatchacceptance

Inspection & quality

Devices

Scheduled and on-demand inspection with real uptime time series; black screen, blur, freeze, noise and signal loss checks with rule-based review.

inspectionVQA

Analytics wall

Reports

Distributions, closure rate, SLA funnel and trends rendered as local SVG; two rate trends only from real inspection runs, never fabricated.

zero-dependency SVGCSV export

Local-first AI

Assistant

Advice with steps, parts and confidence; device risk ranking; local Q&A; repair wizard. Local rules by default, optional LLM only for wording on redacted text.

no external callsexplainable

Asset map

Location

Offline schematic map with pulsing fault markers; optional satellite base map falls back automatically when unavailable. No key shipped in the front end.

offlinefallback

Verifiable audit

Trust

Each audit entry carries seq / prev / hash. Recalculation locates edits, deletions, insertions and reordering; an anchor parameter detects whole-file rewrites.

SHA-256anchor
Boundary: the local hash chain detects tampering but cannot stop someone with write access from recomputing the whole chain. Production requires anchoring the daily head to independent read-only storage and six-month retention.
Architecture

Clear layers, honest attribution.

Operations layerIn-house
cockpitalarm clusteringticketsinspectionVQA rulesassetsanalyticsaudit chainRBAC
Video accessUpstream · MIT
wvp-GB28181-proZLMediaKitGB28181-2016SIP / RTP / RTSP
RuntimeBase
Node.js ≥ 18NginxDocker or systemdTLS 1.2 / 1.3
OptionalOff by default
Tencent Map (own key)LLM gatewayONVIF (licensed separately)
In-house code Upstream open source, original licenses retained Runtime dependency or opt-in

State the boundary before the capability

These are delivery facts, not marketing disclaimers.

  • Built on an upstream: video access uses wvp-GB28181-pro (MIT, © 2020 swwhaha); copyright and license notices are retained, with no implied endorsement.
  • ZLMediaKit notice: its license includes a supplementary term; the “ZLMediaKit” mark in title / Server / User-Agent must not be removed.
  • ONVIF is not open source: it is outside the MIT grant and requires separate licensing, or use GB28181 access instead.
  • AI is local by default: no external requests, no video frames or images. If an LLM is enabled, only redacted text is sent and numeric output is verified.
  • Demo data is labelled: prototype and drill data carry a demo flag and never enter production state or reports.
  • No compliance certification claimed: the security posture is an engineering self-assessment; MLPS level 3 conclusions require an accredited evaluator.
Full license and gap list Read the compliance details →
FAQ

On ownership, data and readiness.

Is DADI-OPS a from-scratch video platform?
No. Video access and streaming rely on wvp-GB28181-pro (MIT) and ZLMediaKit. DADI-OPS builds the operations layer above it: alarm clustering, tickets, inspection, quality checks, assets, analytics, local AI rules and audit. We do not count upstream capability as in-house.
Can my existing GB28181 cameras connect?
Devices compliant with GB28181-2016 can register and stream through the upstream access layer. Site work is required: SIP server ID, domain, port, password, streaming address, and firewall rules. A diagnostic script reports each step from login to device, channel, alarm, subscription and uptime.
Does the AI upload footage or device details?
Not by default. Without an LLM configured, everything runs in local rules with zero external requests and no image processing. If enabled, only redacted text is sent (device names become “site#N”); IPs, GB IDs, personnel and images are excluded, and numbers in answers must match local facts.
Is it ready for large-scale production?
The current release is a prototype and pilot delivery. The operations loop, authentication, analytics and deployment scripts work, but default device and alarm data is illustrative. MLPS assessment, independent audit storage, storage encryption and off-site backup must be completed per site. We recommend a bounded pilot first.

Start with one device and one alarm.

Tell us your device scale, vendors and current access method. We will scope a pilot, assess integration and provide an acceptance checklist.