Facilities AI agent · Alarm and access
Physical Security Monitoring AI Agent
Couple an alarm to the access record and the footage that covers it, hold the candidate event open until an officer dispositions it, and report silence as silence.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named officer
- No auto-closure
- After launch
- Agent Care
What this agent does
Assembles the candidate event, never the disposition
In
- 01
An alarm raises, and the agent draws the access record and the footage covering that door and minute.
- 02
A camera goes dark, and the gap is reported as a gap rather than left to read as a quiet night.
Reason
- 03
A door is forced, and the alert stays open until a named officer dispositions it or hands it on.
- 04
An alert ages without a review, and it is escalated rather than allowed to close itself.
- 05
A badge opens one door twice with nobody leaving between, and that door is marked for a tailgating read.
Decide
- 06
A door is held open beyond the configured interval, and the propped-door pattern is raised against it.
- 07
A recording nears the end of its retention window while its event is open, and the agent says so.
Out
- 08
A credential appears at two sites too close together to travel between, and an officer is given both.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent correlates the signals and holds the candidate event. A named security officer dispositions the alarm, and only that officer closes it.
Example workflow
One alarm, signal to disposition
- 1Alarm signal receivedPanel alarms, access transactions, camera health or guard-tour feeds
- 2Event context assembledThe door, the site it sits on, the credential presented and the minute the signal was raised
- 3Candidate event draftedThe alarm, the access record beside it, the footage found and completeness
- 4Controls appliedTime-window checks, camera-health checks, retention checks and completeness confidence
No human action required
Stages 1 to 4 run unaided, and nothing is dispositioned at any of them — the agent is correlating, and the control-room lane opens at the completeness gate.
Goes to the duty security officer to disposition.
Adds a control-room supervisor read first.
Control-room review
The event is held with its signals, its footage window and the door it sits on.
Disposition · Append evidence · Send to control room
Dispositioned — by a named security officer
- 6Alarm and case records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedCorrelation accuracy, footage coverage, reviewer corrections and what the read found
Corrections
Each control-room correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Dispositioning an alarm as false or as actual.
- Closing an alert because nobody reviewed it.
- Naming the person behind a badge as a conclusion.
- Running recognition outside your configured policy.
Automation boundary
Agent acts unaided
- ✓Correlate each alarm with the access record and the footage window.
- ✓Hold the candidate event open until a named officer.
- ✓Mark footage nearing the end of its retention while.
- ✓Report a dark camera as a gap in the record, never as a quiet door.
Nothing is dispositioned or closed except by a named person, inside the agreed boundaries.
- Judging whether a person tailgated a door.
- Telling a site the estate was secure overnight.
- Setting the priority a panel arms a zone at.
- Changes to access rights, doors or schedules.
Example output
One alarm, annotated
Our retail loss-prevention agent works shrink across a trading floor; this record is what one forced door carried into review.
Candidate event · single alarm
Illustrative example- Alarm
- Door forced, perimeter entrance
- Recorded as
- Correlated with an access record and footage
- Signal
- Door forced
- Evidence of record
- Panel alarm, 3 August 2026
- Confidence
- Held undispositioned
- Held for
- The duty security officer, by name
As received
Taken from the panel log and the access record — it reaches as far as those sources do.
What the record holds
- Panel alarm log
- Access record
- Footage reference
Why no disposition here
Whether the alarm was an actual entry is an officer call, not a model output.
What the score decides
Below the configured threshold an event gets a supervisor read before the officer sees it.
Action
DispositionAppend evidenceSend to control roomValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every alarm
From the door that raised it
Where the evidence is used
Correlation is not identification: the agent holds no biometric and runs no recognition unless your configured policy provides for it, and it names nobody as a conclusion.
An alarm is not an event
A panel signal becomes an event only when a named officer says it is one, and the agent assembles the correlation while that call stays with the officer.
What was checked, and not found
No regime sets a disposition clock for a forced door on a private estate, no public register of alarm outcomes was located, no retention period is fixed for us to quote, and every limit on this page is one your own configured policy sets.
Build an evidence trail
The alarm, the footage covering it and the officer who dispositioned stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Alarm and intrusion sources
- Panels
- intrusion feeds
- Door and zone alarm signals
Access control
- Readers
- credential records
- Badge and door transactions
Video and camera health
- VMS
- camera heartbeat
- Footage index and retention
Agent
Physical security monitoring
- Reads the signals
- Correlates the event
- Holds for the officer
Case and dispatch systems
- Guard tour
- ticketing
- Disposition and case records
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six cordons between the model and the officer
Six cordons drawn one inside the next. Whatever crosses all of them is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an event may stand
- L3
- Puts the disposition in a person's hands
- L4 – L5
- Keep the alarm and the footage behind it
- L6
- Steps back to alert assembly when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the candidate event that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the event an officer reads
- E1Final-output evaluationDid the event record what the alarm actually showed?
- E2Step-level evaluationDid the agent read the right door, the right minute and the live device list?
- E3Tool evaluationDid it read and write the correct alarm record and the correct case?
- E4Confidence calibrationDo low-confidence events actually attract more control-room corrections?
- E5Slice evaluationHow does performance change across specific site types?
- E6Business outcomeHow many events needed a correction before the officer dispositioned?
Floor — the record the estate answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each pinned to the stage where it first shows.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- KP-03
Stale device list read
The camera list read is not the one now in service.
Stage gathers
The alarms, the badges, the doors and the times
02 · Reasoning
2 modes
- KP-04
Bound to the wrong door
A signal is tied to a door it never came from.
- KP-06
Silence read as all clear
A dark camera is worked as a quiet night.
Stage proposes
The alarms, their records and completeness
03 · Tool / write
2 modes
- KP-02
Thin event passed forward
An event moves on without the supervisor read.
- KP-05
Filed against the wrong site
An alarm lands on a building it did not touch.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- KP-01
Dispositioned, evidence unrecorded
The record shows a disposition but not what supported it.
Stage returns
The event an officer dispositions and audit reads
05 · Change / Version
1 mode
- KP-07
Silent correlation regression
A configuration change moves the window, not the alert.
Stage tracks
Model, prompt, correlation rules and event fields
- Sev-1 · an alarm closed with nobody reading it
- Sev-2 · wrong footage reaches the event
- Sev-3 · source degrades, alert held open
Affected slices
Perimeter doors absorb the corrections
A site-level disposition-latency figure can read clean while perimeter doors after hours carry most of the rework. Nestack reports the correction rate by site type, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Perimeter doors after hours | 7.9% | 3.6× | Review | |
| Loading bays and yard gates | 5.6% | 2.6× | Review | |
| Shared-tenancy floors | 3.5% | 1.6× | Watch | |
| Head-office reception | 1.4% | 0.6× | Normal |
Bar: correction-rate lift vs. head-office reception baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a swallowed alarm costs
A loop ends when the swallowed alarm has become a case the next release must pass. That suite is what the next alert raised is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on perimeter doors after hours.
- 02
Diagnose
The door-forced alert that closed itself at three in the morning is read back until one cause remains.
- 03
Improve
Number the change; the alarms that drove it are filed underneath it.
- 04
Verify
Each touched alarm case is run once more, and one red holds it back.
- 05
Learn
It stays on as a standing test, and the retention rules travel with it.
Learn → Detect
The return edge. The next alert is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, event correlation, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Disposition-policy and automation-boundary work.Week 1
- 02Alarm, access and camera sources.Week 1, Week 2
- 03Alarm-to-footage and retention-window rule mapping.Week 1, Week 2
- 04Alarm and access ingestion.Week 2
- 05Alarm, door and footage binding.Week 2, Week 3
- 06Completeness scoring and review routing.Week 3
- 07Officer disposition workflow.Week 3, Week 4
- 08Video-system integration.Week 5
- 09Correlation and retention cases.Week 4, Week 5
- 10Guardrails and disposition controls.Week 4
- 11Alarm-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne site, one year | ProductionProduction monitoring workflow | AdvancedMultiple sites / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Event correlation to your rules | ✓ | ✓ | ✓ |
| Security officer release | ✓ | ✓ | ✓ |
| Device-coverage baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by site | — | ✓ | ✓ |
| Disposition workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Access-and-camera integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-site door schedules | — | — | ✓ |
| Cross-site event packs | — | — | ✓ |
| Large door estates | — | — | ✓ |
| Multi-site retention controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, alarm volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your sites and the doors monitored at each
→Site mapping and alarm capture
Week 102Representative alarm, access and footage records
→Record binding, correlation logic and the event baseline
Week 203Your configured policy on recognition and retention
→Correlation mapping, door binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Alarm, access and video-source assessment, then integration setup
Week 205Alarms you would not want reopened
→Correlation cases and the evaluation suite
Week 406What no camera record may guarantee
→Completeness scoring, review routing, guardrails and release controls
Week 307A named security officer to disposition
→Disposition workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
The spans below are honest working weeks rather than layout, so two of them have to share week five.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Monitoring workflow discovery, door mapping and the automation boundary
W2Source integration and the device-coverage baseline
W3Event correlation, retention logic and release controls
W4Evaluation suite, correlation cases and failure-mode testing
W5Video integration, pilot alarms and targeted corrections
W6One monitoring year run under the security manager, then Agent Care handover
- Reading the band
- Each bar covers only the weeks its own work is named for. Week five holds a pair because the work does.
- At the end of W6
- When the alarm record validates, Agent Care picks the agent up.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Facilities AI agent
Build a monitoring agent around the alert that closed itself while the estate slept.
Bring us one alarm and the record it left behind. If an alert on your estate can age out unreviewed, then a disposition nobody made is the price of every quiet night. The agent holds no biometric and runs no recognition unless your configured policy provides for it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your door estate
- Named officer
- No auto-closure
- After launch
- Agent Care available
Nestack Agents · Alarm monitoring
AGT-AF-08 · Agent Care available after launch