01
Perception
Vision Models
Detect, track and structure the visual signals that matter across people, objects, movement and events.
- Detection
- Tracking
- Evaluation
VLM agent platform
PLATFORM / 01 Vision intelligence that can operate in the real world.
Deqa combines visual models, VLM reasoning, an AI Agent Harness, customer SOPs and edge compute so video becomes governed evidence and action.
One governed stack
From perception to execution — without the black box.
Four connected layers make the AI Agent useful, controllable and ready for real customer operations.
02
Multimodal context
VLM Reasoning
Understands scenes over time by combining video, language, site context and the customer’s operating question.
- Video
- Language
- Context
03
Governed orchestration
AI Agent Harness
Connects the VLM to tools, memory, policies and enterprise APIs so reasoning becomes safe, observable action.
- Tools
- Memory
- Guardrails
04
Executable operations
SOP Engine
Turns customer SOPs into triggers, permissions, escalation paths and the evidence required to close every loop.
- SOPs
- Escalation
- Evidence
DEQA BOX / EDGE AI APPLIANCE
Put the AI Agent where the video lives.
Deqa Box is the secure edge runtime for visual models, VLM reasoning and the local side of the Agent Harness. It keeps critical workflows fast and resilient while respecting privacy and bandwidth constraints.
Near-camera inference
Offline-ready
Secure evidence buffer
Hybrid orchestration
Explore the operating stack
System architecture
From any camera to an SOP-governed action.
A transparent operating stack designed for bandwidth, latency, privacy, permissions and enterprise integration constraints.
Visual Inputs
CCTV · Drones · Dashcams · Mobile
Deqa Box
Edge inference · Secure buffer · Local runtime
VLM + Agent Harness
Context · Tools · Memory · Guardrails
SOP Action
Decisions · Escalation · Evidence · APIs
SOP-governed actions
Low-latency decisions
Enterprise integrations
Deployment choice
Put intelligence where the environment requires it.
Near-camera inference for low latency and limited bandwidth.
Private operation inside the customer network.
Central management for distributed multi-site operations.
Real-time edge action with governed cloud aggregation.
Physical environments
One platform, configured to the question.
Retail & customer spaces
Journeys · queues · dwell · conversion
Warehouses & logistics
Flow · loading · dwell · evidence
Manufacturing & assets
Safety · quality · assets · SOPs
Field operations & infrastructure
Attendance · tasks · roads · sites
Start with one visible problem