01
Incident priority
Severity ranked by sensor data, occupancy, asset value, SLA pressure and historical impact.
// Command & Control · Agentic dispatch
AI proposes. Operators approve.
When an incident is detected, Innfini analyzes the situation, assesses impact, recommends the right SOP, identifies the nearest available responders and proposes the next-best action — all in seconds, with cited reasoning. The operator stays in control. Nothing dispatches without approval (unless a rule explicitly authorizes auto-action).
<3s
Detect → recommend
100%
Explainable + audited
Human
In the loop
Detect
Recommend
Approve
Act
// The screens
Six views. One governed loop.
Each Innfini screen below is an illustrative scenario — a riverside city, a university mall, a port campus — showing the same agentic loop from the first signal to the closed record.
01 · The incident console
Map left. Recommendation right. Nothing moves until a human says so.
The operator opens INC-0007 — a high-severity fire alarm in Zone 4 of a riverside mall — and sees the live map on the left and the end-to-end recommendation on the right: dispatch Engine-12 via the proposed route to Gate 3. The evidence the recommendation rests on is listed and clickable, the available responders are ranked by ETA, and the status line says exactly where things stand: awaiting operator approval, no dispatch issued.
- Recommended action with the proposed route drawn on the map
- Cited evidence · SMK-4-08 smoke detected, CAM-4-12 haze confirmed, EVAC-2 procedure matched
- Responders with live availability and ETA · Engine-12 3 min, Patrol-22 2 min
- Approve dispatch or Escalate — actions execute only after recorded approval
02 · Operational context
Ten signals behind one recommendation.
A recommendation is only as good as what it was allowed to see. For a fire alarm at the main entrance of Riverside Commons, Innfini assembles the full operational context before it proposes anything: the priority, who should be assigned, how it escalates, which SOP applies, what resources are nearby, who has to approve, what to communicate, the likely impact, the SLA at risk, and how similar incidents ended.
- Sensor evidence, available responders and previous outcomes surfaced alongside the map
- Units A1 and B3 shown with distance, exits and first-aid points marked
- Every one of the ten signals opens to its source
- The same context drives the confidence value the operator sees
03 · The loop, end to end
Detect, recommend, approve, act, learn.
A crowd disturbance at University Mall shows the whole loop on one screen. Signals from three cameras are correlated into one incident; the agent proposes a Community Response unit with a 3-minute route and cites its evidence; the operator approves at the console; the approved dispatch rolls; and when the incident is resolved the outcome is reviewed, the citations are archived and the model is updated for next time.
- Live layers — roads, transit, parking, cameras, crowd analytics, EV charging, incidents, weather
- Recommendation with citations · Cam 12, 14, 27
- Operator decision recorded before any action
- Outcome review closes the loop and improves the next recommendation
04 · Action policies
Right action, right people, safer outcomes.
Not every action deserves the same gate. Policies decide, per action type, how much autonomy the agent has: AUTO for pre-approved routine tasks such as sensor resets and status updates; ATTENDED for anything that needs an operator to approve, such as a maintenance dispatch or a variable message sign; SUPERVISED where an operator and a supervisor must both sign, such as lane closures or multi-agency coordination; and BLOCKED for actions that are human-only, such as emergency-services dispatch and system changes.
- Four policy levels, configurable per SOP, per role, per severity
- A road hazard on Riverside Drive sits at ATTENDED — approve, modify or escalate
- Every decision recorded with identity and time
- Blocked actions never automate: the AI advises, people execute
05 · After approval
The decision history writes itself.
Once the operator approves, the record keeps going. Alarm detected at 18:36, plan proposed at 18:37, operator approved at 18:38, Engine-12 dispatched at 18:39 — with the sensor source, the approval and the crew dispatch each linked to their evidence. The response is in progress, the next step is field confirmation, and the after-action report is already most of the way written.
- Timeline of every step with its timestamp
- Sensor source linked · mall exterior camera
- Approval recorded at the operator console
- Crew dispatched and tracked en route to Gate 3
06 · The same loop on a port
Not only a city product — a terminal, a warehouse, a campus.
The agentic loop is the same wherever Innfini runs. On a port logistics campus a fire alarm in Building 7, Zone 4 is confirmed from the alarm and the detector, Response-12 is suggested from the fire-vehicle staging area with a 3-minute ETA and procedure EVAC-2, and the decision waits for the operator. Evidence reviewed, plan proposed, approval pending — on a port map instead of a city map.
- Building-level zones on the terminal with the alarm pinpointed
- Suggested unit, availability, ETA and procedure in one card
- Three-step status · evidence reviewed, plan proposed, approval pending
- Site overview keeps the wider terminal in view
// The agentic-dispatch loop
Five stages. Always closed.
01 · Detect
Surface the event
Smoke, camera, RFID, sensor or operator report — every channel routes into the incident graph in real time.
02 · Recommend
AI proposes
Severity, SOP, responders, ETA, escalation path, similar past incidents — generated with citations and confidence.
03 · Approve
Human gate
The operator stays in control. Approve, escalate, modify or override — every decision recorded with identity + time.
04 · Act
Execute
Dispatch, alert, broadcast, lock-down, notify — through the integration layer. Outcomes streamed back to the canvas.
05 · Learn
Improve
Outcome feeds the model. Next time the same pattern arrives, recommendation quality and confidence improve.
// What the AI recommends
Ten signals. One recommendation.
Every recommendation is a fusion of operational signals — not a black-box "best guess".
02
Team assignment
Nearest available crew with the right skill, equipment and certification — plus an ETA.
03
Escalation path
If unresolved in X minutes, escalate to Y — multi-level paths defined per SOP.
04
SOP checklist
The right SOP attached, with required steps prefilled and trackable in-line.
05
Nearby resources
Cameras, vehicles, suppression systems, first-aid stations and exits within the incident radius.
06
Required approvals
Threshold-gated approvals — by cost, severity, role or zone — surfaced inline.
07
Comms actions
PA, SMS, push, signage, mobile alert — scripted per SOP, ready to launch with one approve.
08
Estimated impact
People affected, asset risk, downtime, financial exposure — quantified up front.
09
SLA risk
Probability of SLA breach with current resourcing — and what to change to recover.
10
Similar past incidents
Two or three precedents with their actions, outcomes and time-to-resolve — for fast pattern-match.
// Governance · operator-in-the-loop
AI proposes. The operator approves.
Innfini is AI-assisted, not AI-autonomous. Approval modes are configurable per SOP, per role, per severity — from full auto-execute on low-risk routine actions, to mandatory dual-control approval on high-stakes ones.
Auto
Routine, low-risk
Pre-approved actions like camera cycling, log queries.
Attended
One-click approve
Operator must approve, but recommendation is presented inline.
Supervised
Dual-control
Operator + supervisor approval required before action executes.
Blocked
Human-only
Some actions never automate — AI advises, humans always execute.
// Why agentic dispatch matters
From paging-on-a-glance to coordinated response.
01 · Faster
−68% Time to Dispatch
From detection to crew rolling — measured across deployed sites.
02 · Smarter
Right SOP, First Time
SOP attached automatically — no manual lookup, no missed steps.
03 · Safer
Human Gate Always
Every dispatch is approved. Every approval is logged with identity + reason.
04 · Auditable
Cited Reasoning
Every recommendation explains itself — sensor IDs, camera frames, past cases.
Agentic Dispatch is the AI-assisted, human-governed response loop at the heart of Innfini Command & Control. The platform proposes. The operator decides. The decision is explained. The outcome is learned.
See agentic dispatch on your operation.
A live walkthrough with our solutions team — your SOPs, your responders, your approval thresholds.