Manifesto: The End of Remote Control
“A Label is Not Courtesy; It is a Declaration.”
The era of controlling home devices one by one through a smartphone is coming to an end.
Matter has lowered the walls and opened a standardized path for control. Now, manufacturers face an era where they can no longer know who is sending control requests to their devices. Not just Google, Amazon, or Apple — but Claude, Grok, ChatGPT, or a personal agent someone built tonight can send commands to your device through a Matter Controller.
In this era, the first thing a manufacturer must do to protect themselves and their users is to make a 'Declaration.'
Until now, labels were treated merely as a domain of UX—designed to make things look pretty and intuitive for human users inside an app. But in an environment where AI controls the physical world, a label is not merely a name tag. It is a technical declaration that the manufacturer leaves for the outside world.
Manufacturers must clearly declare the following inside the device, in advance:
- What is this endpoint?
- What does this action mean in the physical world?
- Which safety boundaries must never be crossed?
- What is the responsible scope of use for this device?
This is not mere metadata. It is a technical constitution — a public declaration by the manufacturer stating: "This is how this device is to be understood, and this is how it is to be used."
If an incident occurs, what the manufacturer explicitly declared may become a critical basis for discussing responsibility. Whether an AI agent was able to read that declaration, and whether it respected those boundaries, will be the defining questions of the physical control era.
This platform provides manufacturers with a technical safeguard, and AI with a precise map for understanding the physical world.
This device does not exist for a single smart home app. Any AI agent connected through Matter may send control requests to this device. But that control is permitted only within the authority of the Label declared by the manufacturer.
A Label is not courtesy. A Label is a declaration.
The First Device AI Can Actually Understand
Every device built so far was designed for humans to operate.
What comes next are devices built for AI to understand. The question is: which devices get there first?
The Problem With How We Build Devices Today
Device types emerged from the way humans understand devices — read by humans, installed by humans, operated by humans. For that era, it was the right structure.
AI doesn't understand devices that way. It understands actions and their boundaries. The twenty-year-old structure isn't wrong. The era just changed.
Traditional IoT development starts with a template. Pick a device type from the platform's list. Fit your hardware into it. Map your clusters. Copy reference firmware and modify.
This works for humans. For AI, it's a dead end.
AI reads the device type and cluster structure, then guesses what the device actually does. It controls a kitchen light and an electric blanket the same way. Safety boundaries are hardcoded inside firmware — invisible from the outside. When AI issues a command, it doesn't know what's allowed. It infers.
Inference is expensive. Inference is uncertain. And when inference controls a physical device, uncertainty becomes a safety problem.
A Different Starting Point
Instead of asking "which device type do I pick?", ask one question first:
What is this action?
Not what hardware it runs on. Not which cluster it maps to. What does it actually do, and what are its limits?
When you answer that question explicitly — with a name, a boundary, and a context — something changes. The action has an owner. The manufacturer owns what the device does and what limits it carries. The user owns what that action means in their life.
Return each fact to its rightful owner, and AI no longer needs to guess.
This is what Nemo & Anna is built on.
How It Works
Answer 9 questions about a single action. The answers are saved as JSON.
The manufacturer's responsibility for functions and limitations is also just one of the nine questions.A firmware runtime reads the JSON and generates a valid Matter device automatically — cluster mapping, commissioning, GPIO conflict protection, state management. No C++ code written.
Note
This is an illustrative example.
There is no actual button or device called “Sunday Jjapagetti.”This JSON becomes a Matter endpoint.
Manufacturer-defined safety information is stored in Fixed Label (0x0040) and can be conveyed to the user.
Action identity is expressed through a structured labeling pattern using User Label (0x0041).
We’ve been raising this issue across major ecosystems:
Very different physical actions are still treated as the same “switch.”
link: https://github.com/project-chip/connectedhomeip/issues/71521This works within existing Matter behavior without requiring specification changes.
When the action definition changes, update the JSON.
No firmware rewrite. No reflashing.
The ESP32-C6-DevKitC-1-N8 is widely available — Mouser, Digi-Key, and distributors worldwide carry it. The barrier to entry is the idea, not the hardware.
What This Means for AI
A device built this way tells AI exactly what it needs to know — without inference.
- What is this action? The name declares it.
- What are the limits? Fixed Label carries them, read-only.
- When is it valid? The user's context declares it.
AI doesn't guess. AI reads, maps, and executes — inside boundaries that have already been declared by their rightful owners.
Devices that declare this structure enter the AI ecosystem immediately. Devices that don't will wait for the platform to catch up, or for the model to get smarter, or for someone to write the right prompt.
Zero-Configuration AI Context Ingestion (How Responsibility is Defined)
You don't need complex cloud APIs to enforce safety boundaries. By leveraging the native Matter protocol, the manufacturer's declaration is automatically injected into the AI's brain before any physical action takes place.
[ESP32-C6 Firmware]
└ Define Fixed Label (Physical Boundaries)
│
▼ (1-Time Native Matter Commissioning)
[Matter Controller DB (Google / Apple / Amazon AI Speakers)]
└ ★ Automatic Ingestion: Manufacturer's Declaration is permanently cached ★
│
▼ (User voice command triggers LLM inference)
[LLM Agent Context Window]
└ Available for injection as System Prompt / Context.If the platform fails to pass this declaration to the AI, liability shifts to the platform, not the manufacturer.
Why This Matters Now
Most companies building AI-controlled physical devices are stuck. The technology works. The liability doesn't. There's no clear answer to "when something goes wrong, who is responsible?"
This structure answers that question before it becomes a problem. The manufacturer declared the boundaries. The user approved the context. AI executed within what was declared. The record exists.
That's what makes it shippable.
The design principles behind this structure:
→ Physical AI Safety: Ownership and Execution Boundaries
Every product comes with a user manual for humans.
But for AI, we have mostly relied on probabilistic guessing.
This project changes that.
Manufacturers must define AI-readable manuals — structured as discrete action units.
Through the Nemo & Anna framework, each device capability is declared in JSON internally, then exposed to the outside world through standard Matter protocol. Each action can carry its own name, purpose, and execution boundaries, allowing AI systems to understand what is allowed before physical control begins.
This removes the need for AI to guess and provides a practical path toward responsible ownership.
Now, it is time for the ecosystem to respond.
We hope to see this safety-first structure expanded across more platforms, so that AI operates only within boundaries defined by manufacturers and approved by users.
Nemo & Anna is not a future proposal. It runs on ESP32-C6 today, within existing Matter infrastructure, without any specification changes.
If you're building a device that AI will control, this structure is ready to apply today.
The hardware is already on the shelf. The framework runs today.
What's left is declaring your device first.
The devices that AI can understand first will define the next ecosystem.
The ecosystem will be defined by whoever declares first.
Try it at anna.software
All firmware, runtime logic, and device specification format are fully open-sourced.
No black box. No hidden boundaries. Everything AI needs to understand this device is readable, auditable, and forkable.
→ github.com/anna-soft/Nemo-Anna
Comments