DATA TALE · AI OPERATING LAYER
From device to AI workflow: an open communication protocol for operational systems
SaveVision shows how devices, live data, domain logic, AI proposals and user interfaces work together over a controlled communication protocol. The concrete use case is remote guidance — the transferable capability is building an open communication layer between device, backend, operator interface, data streams and AI logic.
The situation
Many AI applications fail not because of the model, but because of the missing connection between user, data, tools, domain logic and operational decision.
SaveVision makes this problem visible through a critical medical use case: a person on site wears smart glasses, provides live context and needs concrete guidance. An expert sees the situation remotely, assesses it and sends back controlled action instructions.
The structural challenge behind it, however, is broader: organisations need communication layers through which multimodal data, approvals, tasks and AI proposals can flow reliably between devices, applications and people.
A single tool is not enough for that. What is needed is an operating layer that captures, distributes and processes context and plays it back into a usable interface.
A real-time communication layer between device, backend, operator and AI
SaveVision connects smart glasses, an iOS app, live video, audio, location data, map information, an operator cockpit, backend events and optional AI proposals into one end-to-end workflow.
The person on site streams their perspective. The system transports video, voice, location, instructions, drawings, reference images and control information over an open communication architecture. The expert works in a cockpit, assesses the situation and sends approved instructions back to the glasses display.
The system logic follows a clear pattern: capture context, hand it over, assess, prepare, approve and play it back. This turns an uncertain live moment into a structured, controllable work process.
The role of AI is deliberately limited: it can generate proposals, but the final decision and approval stay with people.
- Live first-person video and two-way audio between the scene and the operator
- Smooth real-time image transmission for the demonstrated remote-guidance workflow
- Multimodal data transport for video, voice, text, location, maps, images, drawings, events and tasks
- Operator cockpit with live image, annotation, mini-map, instructions, session logic and approval controls
- Smart-glasses output with text banners, visual markings, reference images and wayfinding
- Open and self-hostable components such as Matrix, MatrixRTC, LiveKit as well as open map and routing services
- Optional AI proposals with human approval instead of automatic delivery
The project originally emerged at START Hack Vienna 2026, where it was awarded in the hardware track. What matters is the working architecture that came out of it: a platform example for open, multimodal real-time communication between device, backend, operator interface and controlled AI logic.
The example has been developed and maintained further since then. It is, however, not to be understood as an approved medical device or a substitute for professional emergency care.
Business value
The business value lies in the ability to make complex operational situations manageable through a controlled communication and action layer.
This pattern is relevant for companies because AI workflows will not consist only of prompt and answer in the future. They have to take in context from different sources, hand it over to models, tools or domain logic, and feed results back into the right working interface.
SaveVision shows what such a layer can look like: with live context, structured data flows, human control, custom interfaces and an architecture that is not fully dependent on closed platforms.
The public demo and the open-source repository make the implementation verifiable.
- Multimodal data is translated into a usable workflow
- Experts or central teams gain better live context for decisions
- AI proposals become reviewable before they turn into operational actions
- User interfaces, backend logic, models and devices work together over defined handovers
- Open and self-hostable components reduce dependence on individual platforms
- Expertise can be translated step by step into repeatable digital process logic
Why this matters
SaveVision is more than an example of remote guidance. The real value lies in turning live context into a controlled digital workflow: capture, transmit, assess, approve and play back.
Exactly this capability becomes decisive for the next generation of AI systems. Companies need not just better models, but robust ways to move context, data, tasks, approvals and results between people, applications and AI systems.
The example shows how AI can be embedded into operational workflows without handing responsibility to the system. Proposals can be prepared, but the professional decision stays traceably with people.
This shifts the focus from individual AI tools to a steerable operating layer: AI becomes part of a process, not a replacement for the process.
Transferable relevance
The transferable capability lies in building open communication and control layers for AI-supported workflows.
This pattern is relevant whenever information from different sources has to be brought together, processed by domain logic or models, and made usable again in a suitable working interface.
In the context of a Knowledge Operating System, SaveVision would be a vivid example of the communication layer: users provide context through an interface or a device, the backend structures and distributes the information, AI or domain logic processes it, and the result is played back in a controlled way.
For companies, this creates more architectural control: custom interfaces, custom data flows, custom approval logic and — where it makes sense — self-hosted or local models can be connected more deliberately.
- Custom chat and working interfaces for controlled AI workflows
- Multimodal data pipelines for text, voice, video, location, images and structured events
- Agentic workflows with clear handovers between user, tools, models, approvals and operational actions
- Remote guidance for field service, maintenance, inspection, technical incidents and safety processes
- Controlled AI infrastructures for organisations with high requirements for data protection, traceability and technical sovereignty
The reusable pattern is: capture context, transport it in a controlled way, process it through domain logic or AI, build in human approval and feed results back into the right working interface.
KEY TAKEAWAY
SaveVision shows how Kernity can build open communication and operating layers for AI-supported workflows: from the device through data streams and domain logic to controlled action in the user interface.
NEXT STEP
A similar initiative?
In the free initial call we clarify whether Kernity is the right partner for your situation – and which entry point would make sense. The goal is not to sell a solution on the spot.
30 minutes · video or on-site · no preparation needed