JBKR designs and operates local-first automation: the house, the building or the yard keeps working without internet and with no door open to the outside. In 2026 it deployed, in an undisclosed home, Home Assistant with 32 Zigbee devices, 24 automations, a software alarm, cameras with local AI people detection and VPN-only remote access, with 27 documented decisions. In 2023 it prototyped a networked LED panel for a logistics terminal. Zigbee since 2022.
Updated on
Alarm: HomeLights: 0/4Gate: closed
Everything here is a local simulation: every tap produces an event on the bus, as in a real house running MQTT.
Someone arrives at the gate: the camera detects the person and the AI describes the scene. With the alarm set to Away, the siren goes off.
Sensors, bulbs, plugs and locks from different brands join the same Zigbee network and publish on the same MQTT bus. The vendor cloud becomes optional.
32 Zigbee devices on a single mesh network, coordinated by a local dongle
Lights that follow presence and switch off by themselves, a siren that respects the alarm mode, a critical phone alert with an action button. Simple rules, tested and documented.
24 automations in operation: lighting, monitoring and alarm
Remote access only through a VPN, two-step login, cameras blocked from reaching the internet, secrets out of the code and backups in more than one place. Security is design, not an add-on.
Cost of the AI that describes alerts: from hundreds of dollars a month to cents, by analysing only what matters
A mini PC in a cupboard runs Home Assistant, the broker, the video detector and the VPN in containers, with a UPS and backups. The same design serves the edge of a company.
27 documented decisions, from hardware to retention rules
Data leaves the house or the building through an API and reaches where the operation needs it: a portal, an ERP, a wall panel or an LED matrix in a truck yard.
Panel prototype with 727 addressable LEDs reading messages from an API
Scale and compliance change; the local-first, secure and documented design stays the same.
Offices, condominiums, clinics and small industry.
Houses and flats that want comfort without depending on a cloud.
Manufacturers, integrators and technical teams that need software and firmware.
A whole automated home and an LED panel in a logistics yard: the same care with network, security and documentation.
An IoT project starts with the network and the risks, not with buying devices.
Network, existing devices, coverage, risks and what must keep working without internet.
Protocols, topology, server, security and backup plan, device list with cost.
Server, coordinator, devices, automations and integrations with what already exists. Electrical work with a partner.
Recorded decisions, dashboards, credentials in the client vault and training for whoever will operate it.
Updates, new devices, rule adjustments and monitoring, on a monthly plan or on demand.
The first criterion of a JBKR automation project is what happens when the internet drops or the manufacturer shuts down its cloud: nothing. The server runs inside the house or the building; devices join a Zigbee mesh network coordinated by a local dongle and publish on the MQTT broker; automations, alarm and dashboards depend on no external service. The vendor cloud becomes optional, and voice assistants stay off by default.
In the 2026 residential project that meant 32 Zigbee devices from three different brands (opening and presence sensors, plugs, bulbs, a siren and an outdoor light) on a single network, with Home Assistant at the centre, Zigbee2MQTT as the bridge and Mosquitto as the bus. Dashboards per room, a wall view for the tablet and an energy dashboard complete the delivery.
Home automation is, first of all, a system that controls doors, gates and cameras. That is why access from outside comes only through a VPN tunnel, with HTTPS inside the tunnel and no port forwarded on the router; administrators use two-step login; the wall-panel user is local and unprivileged; repeated attempts are banned; server access is by key. Cameras and the recorder are blocked from reaching the internet.
Secrets (Zigbee network keys, tokens, passwords) stay out of the code and the versioned files; backups go to the local disk and to two destinations outside the house. The alarm is software: modes, entry and exit delays, a physical remote and critical phone alerts with an action button. The security and privacy page describes how JBKR handles access and data as a processor.
Recording everything all the time costs disk and helps nobody: in the 2026 project, two days of continuous recording produced 287 GB. The answer was a local people detector running on the mini PC's integrated GPU, at about 8 ms per analysed frame, face recognition to tell residents from visitors, and a retention policy defined after measuring, not before.
Every person alert gets a short description written by a generative-AI model ("visitor standing at the gate, no vehicle"), so the phone notification can be read in a second. The first version analysed everything and cost hundreds of dollars a month; the version in operation analyses only what matters and costs cents. JBKR's responsible-AI principles apply here: declared use, local processing whenever possible and human review.
The server is an Intel mini PC in a cupboard, on a UPS: Home Assistant OS, Zigbee2MQTT, Mosquitto, the video detector and the VPN run as managed containers. The project was born on a Raspberry Pi and migrated to the mini PC without losing history, automations or configuration, with the migration plan documented step by step.
Every relevant choice (hardware, alarm, detector, retention policy, remote access, backups) lives in a short decision record with date and reason: 27 in the 2026 project. It is what lets the owner understand their own house two years from now, and it is the same design JBKR uses for the edge of a company: process video and data on site and send up only the summary.
Home Assistant exposes REST and WebSocket APIs; JBKR uses both to generate dashboards by script (versioned as code), read states and trigger actions from other systems. For companies, that is how sensor data reaches the ERP, the portal or the reception, and how the gate, the turnstile or the alarm obey business rules.
At the edge, hardware: in 2023 JBKR prototyped, for a logistics terminal, a message panel with 727 addressable LEDs, driven by an Arduino with Ethernet that queries an HTTP API with a token and draws the text with a custom font. The same path (Arduino or ESP32, HTTP or MQTT, a token, EEPROM for local configuration) serves industrial sensors, kiosks and yard panels.
It starts with the assessment, not the purchase: network, coverage, existing devices, risks and what must keep working without internet. From there come the design (protocols, topology, server, security and backup plan, device list with cost), the deployment and integration with what already exists, the documentation with recorded decisions and the training of whoever will operate it.
JBKR is accountable for the software, the network, the security and the integration; electrical and physical installation stay with a partner electrician or installer, or with the client's professional, guided by us. After delivery, support (updates, new devices, rule adjustments) runs on the monthly plan or on demand.
| Platform or protocol | What for | Status at JBKR |
|---|---|---|
| Home Assistant OS | Automation hub: devices, automations, dashboards, APIs | In operation (2026) |
| Zigbee2MQTT + Mosquitto (MQTT) | Zigbee mesh network and local event bus | In operation (2026) |
| Alarmo | Software alarm with modes, delays and a physical remote | In operation (2026) |
| Frigate + OpenVINO | People detection and face recognition on the integrated GPU | In operation (2026) |
| Tailscale (VPN) | Remote access with no open port, HTTPS inside the tunnel | In operation (2026) |
| Arduino, ESP32, Ethernet, WS2812B | Hardware at the edge: sensors, panels, prototypes | Prototype delivered (2023) |
| ESPHome, Tasmota, Matter | Local firmware for Wi-Fi devices and the Matter standard | Capability |
| Modbus, LoRa, Z-Wave | Industrial sensors, long range and Z-Wave networks | On demand, with specialists |
| Model | How it works | Best for |
|---|---|---|
| Fixed-scope project | Scope, deadline and price defined in a written proposal | Websites, apps, integrations and MVPs with clear scope |
| Monthly plan (retainer) | Monthly hours with priority support: Care or Expert plan | Support, continuous evolution and systems in production |
| Hourly (On-Demand plan) | From BRL 250/hour, estimate before the work starts | One-off fixes, assessments and proofs of concept |
Same-business-day response · Critical-failure fix started within 24 hours on monthly plans
Yes, that is the design criterion. The server, the Zigbee coordinator, the MQTT broker and the automations run inside the house or the building; if the internet drops, lights, alarm, sensors and dashboards keep working. Only remote access and phone alerts depend on the connection.
In most cases, no. Zigbee devices from different brands join the same local network; Wi-Fi bulbs and plugs can be integrated or, when the manufacturer allows it, reflashed to speak MQTT; cameras with RTSP go into the local detector. Voice assistants are optional and stay off by default in privacy-focused projects.
It is, as long as there is no door open to the internet. JBKR publishes remote access only through a VPN tunnel, with HTTPS inside the tunnel, two-step login for administrators and a ban after failed attempts; the cameras are blocked from reaching the internet at the router. What we never do: forward ports, keep default passwords or expose the dashboard on a public address.
A home with a local server, a Zigbee coordinator, a software alarm and AI cameras is a fixed-scope project of weeks, plus the devices, bought by the client from the list we deliver. A building or a plant with ERP integration is a project in stages. The initial assessment defines the list and the price. One-off work uses the On-Demand plan (from BRL 250/hour, estimate before the work starts); continuous evolution and support use the monthly plans: Care from BRL 1,125/month and Expert from BRL 2,050/month. Every proposal is written and itemized.
JBKR is accountable for the software, the network, the security and the integration. When there is electrical work (new points, panel, camera cabling), we work with a partner electrician or installer, or with the client's trusted professional, and deliver the project with the points marked.
Yes. Design, server configuration, automations, security and support are remote; the physical installation stays with a local partner or the client's team, guided by us over video and documentation. In Curitiba and the surrounding region we do the assessment visit in person.
The same design (mini PC, containers, MQTT broker, local video detector, VPN, backup) is what is called edge computing: process video and data on site and send up only the summary. It serves receptions, yards, shops and small plants that need to react in seconds and cannot depend on the cloud.
Cameras produce personal data. We design with short, justified retention, local processing (images never leave the client's server), monitoring notices where the law requires them, and restricted, audited access. Face recognition only with a legal basis defined by the client, and never in public areas. The security and privacy page describes JBKR's role as a processor.
Senior Solutions Architect | AI & Machine Learning Specialist
Jean C Becker · founder of JBKR, creator of AironCore and Apicio, in technology since 2006. See the track record →
In a free 30-minute conversation we map network, devices and risks. You leave with a proposed assessment, timeline and price in writing.
