Move with the operation
The rack is not always coming with you.
Edge brings the complete local stack to on-scene, vehicle-borne, and single-station deployments.
OpCopilot Edge
OpCopilot, at the surveillance station.
A compact AI appliance that sits directly at the station — close to the hardware, zero latency, nothing leaving the site. The same agent team and the same ConOps reasoning, on a box that can go in a vehicle.
Shipping · customer-deployed · human-in-the-loop
Edge node / site 01
Local stack active
Why an appliance
Reasoning where the operation meets the signal.
The operator furthest from the data centre is often the person who needs the intelligence layer first.
Keep the boundary local
The data often cannot leave the site.
Feeds, models, doctrine, and operational memory remain inside the customer-controlled deployment boundary.
Start with one station
Prove the workflow before scaling it.
Edge is the way most command-centre teams start with AI-assisted operations.
Platform
The full stack, not a thin client.
Edge runs the OpCopilot reasoning layer locally and does not call back to a Server instance.
- Compute
- GB10 Grace Blackwell appliance with 128 GB of unified memory.
- Form
- Ruggedizable for on-scene, vehicle-borne, and single-station deployment.
- Model stack
- Runs the full local model stack inside the deployment boundary.
- Availability
- Edge is running today and available to demonstrate on the hardware.
What runs on it
The same operational capability as every edition.
Moving reasoning to the station does not change the skills, connectors, or approval flow.
Skills
Your ConOps, encoded.
The same versioned doctrine and operational skills run on Edge, Server, and Plugin.
Connectors
Station systems, connected.
MCP connectors bring sensor feeds and operational systems into the local agent team's awareness.
Control
The operator still decides.
Recommendations remain traceable and actions require human approval.
Demonstrable now
See the agent team running on the appliance.
The demonstration follows live event ingestion, ConOps reasoning, operator review, and report generation.
Feeds stay local
Station data is ingested inside the customer-controlled boundary.
Reasoning stays complete
Edge runs the full agent and local model stack.
Decisions stay human
The operator reviews sources and approves each action.
Read the silent demo transcript
00:00–00:20 / Sign-in and guardrails. An operator signs in, asks what the assistant can do, and receives a summary of maritime awareness, operational calculations, search-and-rescue support, weather, reporting, and information-fusion capabilities. The assistant states that it will not make command decisions, interpret rules of engagement, or task assets autonomously.
00:20–00:35 / Operational source material. The document workspace shows operational files available to the assistant, followed by a real-time event table summarising incoming data and processing status.
00:35–00:55 / Live feed awareness. A maritime room receives successive radar-tracking insights for Target 001. The insight rail preserves each timestamped source event as the reported range decreases.
00:55–01:30 / Correlation and report drafting. The assistant reasons over the track updates and drafts a structured SITREP covering the situation, track history, threat assessment, missing environmental and asset data, outstanding actions, and decisions requiring operator authorisation.
01:30–02:05 / Updated operational picture. The event-processing view shows new insights and recommended checks as the target continues closing. The operator requests a SITREP from the latest insight; the resulting report records confidence, CPA/TCPA uncertainty, observed changes, an urgent close-quarters risk, data gaps, and actions for operator review.
OpCopilot Edge and OpCopilot Server were presented at the World Border Security Congress 2026.
ShippingOpCopilot editions