Skip to content

Response and agents

One engine, from event to response.

Compose flows on a visual canvas from twelve node types, drop agents into the graph and run them on a durable engine on Temporal that retries, routes errors and rolls back. Omnipercept adds the video and analytics nodes; every run is recorded node by node.

Omnipercept / Response flowRUN-0281 · 20:14:16
North entrance · Rule v3Completed
Event
Person in restricted area01 · response-0
Rule
After-hours access02 · response-1
Notification
Email · webhook · Telegram03 · response-2
Unit assigned
Unit 02 · Shift B04 · response-3
Ticket closed
TKT-1844 · Linked evidence05 · response-4
ALT-3104 → TKT-1844 · Acceso norte · Unidad 02

ALT-3104 → TKT-1844 · North entrance · Unit 02

Demo · Real product interface, live in your browser. The procedure and its data are an example configuration.

The canvas

Twelve node types on a visual canvas.

Event, condition, wait, human approval, webhook, ticket, call and an agent under policies. Omnipercept adds the video and analytics nodes: find the camera clip and seal the event evidence.

Wire them on a canvas, with no orchestration code and no glue services to maintain.

flow · canvas
Event
Condition
Find clip
Seal evidence
Notify
Call
Ticket
Human approval
Webhook
Wait
Agent
End

Agents inside the flow

Drop an agent into the graph.

A mission node works on the canvas itself as the run proceeds: it reviews the clip, confirms the people, seals the evidence and drafts the notice. The flow moves on to the call and the ticket only when the agent returns its result.

It is the same governed agent as in the console, with the same tools and policies, now as a step in your procedure.

Camera event

Hands up · Island 2

Verify alert
verify-event
operator-agent
Done11 · $0.19
4/4 · 100%
  • Review clip
  • Confirm people
  • Seal evidence
  • Draft notice

Call the center

Done

Open ticket

Done

Demo · Real product mission node, live in your browser. The event, agent, turns and cost are an example configuration.

Durable by design

Runs on Temporal. Built to recover.

The engine executes on Temporal: it retries each step under its policy, routes errors to an error output and rolls back with compensation when a step must be undone.

Every published version keeps a snapshot. Review what changed between versions and return to the previous one in a single action.

Runs on Temporal. Compensation built in.

retry policy error output compensation version rollback
Live · v3 Draft · v4

Watch it run

See the run land, node by node.

The live canvas lights its edges by node status as the run proceeds. Every run is recorded node by node, with its timing, inputs and outputs.

RUN TIMELINE
run · v3 · #1187
nodestatusms
event · hands up done 4
video · find clip done 212
condition · schedule done 9
agent · verify alert running …
call · centre queued …

Demo · Run record with example data; nodes and timings are an example configuration.

Live canvas
Edges change with node status: done, running, queued or failed.
Time per node
The timeline shows how long each step took and where the run stopped.
Inputs and outputs
Review what each node received and returned, with the event clip linked.

Triggers

Start it however operations start.

A leader-elected cron, an authenticated webhook, a record change with a filter, a person who asks for it, or a camera detection. The trigger is a node like any other.

Cron

Leader-elected schedule.

Webhook

Authenticated inbound from another system.

Record change

Event filtered by the chosen condition.

Manual

On demand, when a person asks for it.

Analytics event

A camera detection starts the flow.

Permission before action

Reads freely. Writes with permission. Everything audited.

Place human approval wherever your procedure requires it. The engine holds the pause until a decision arrives; the agent works within agreed tools and policies.

A concrete proposal
The request shows the action and the context behind it.
An explicit decision
A person approves or denies before the flow continues.
A recorded decision
The request, permission and result remain linked.
9:41
Ops agent Online
I drafted the end-of-shift summary email to the ops team.
Ready to send. This one needs your approval.
send_email executed

Approved

send_email · confirmed on device

Demo · Real approval interface with an example request. Actions and permissions are configured for your operation.

What gets enabled

Scope

Tools and policies define what each agent can read or execute. Engine outbound telephony is configured with the provider and validated in the pilot.

Capabilities · Response and agents
  • Alert rulesIncluded
  • Alert inboxIncluded
  • Alert deduplicationIncluded
  • Notification delivery trackingIncluded
  • Camera preset actionConnects
  • Relay activationConnects
  • Response workflowsIncluded
  • Approval checkpointsIncluded
  • AI operatorOptional
  • Assistant memoryOptional
  • Specialized assistantsOptional
  • Agent toolsOptional
  • Agent permissionsOptional
  • Per user agent exceptionsRoadmap
  • Scene descriptionsOptional
  • Business questionsOptional
  • Procedure assistantOptional
  • Assistant budgetOptional
  • Automated callsConnects
  • Business messagingRoadmap
  • Reads flow, writes waitIncluded
  • Plan modeIncluded
  • Fail-closed policiesIncluded
  • One surface for people and agentsIncluded
  • Your choice of modelsOptional
  • Agents evaluated before useIncluded
  • No-code canvasIncluded
  • Flow triggersIncluded
  • Video and analytics nodesIncluded
  • Retries and compensationIncluded
  • Human approval in the flowIncluded
  • Per-node run recordIncluded
  • Versions and rollbackIncluded
  • Engine outbound telephonyIncluded
Agreed procedure
Define the event, owner and steps before activating the flow.
Scoped tools
Enable only the queries and actions each agent needs.
Write with permission
Actions requiring approval wait for an explicit decision.

Before expanding

Judgment stays in your hands.

Plan mode
Review the proposal before allowing execution.
Closed policies
If authorization is missing, the action stops.
Decision record
The request and its outcome are kept for review.

A pilot at your site

Test a procedure before automating it.

Choose an event, an action and its approver. The pilot walks through that procedure with your team.