Frequently Asked Questions

The questions that land in our inbox most often, answered plainly. Still stuck? Email us — a real engineer replies.

Does the Bobulator really hit sub-8-millisecond latency?

Yes, and it’s a worst-case number, not a median. The deterministic real-time path — normalize, decide, schedule transmit — runs on a dedicated Cortex-M7 at a fixed 1 kHz tick, isolated from the web stack on the ESP32. A slow API request literally cannot delay a decision. See the per-layer budgets in the architecture.

Can it run all three radios at the same time?

Yes. BLE 5.3, Wi-Fi 6, and LoRa each have independent antennas and run concurrently. Typically LoRa carries long-haul, low-rate traffic while BLE and Wi-Fi handle the low-latency path — but you route any decision to any radio with a one-line rule.

What happens when the internet goes down?

Every decision is made at the edge, so the device keeps running. In a mesh or hybrid topology, only cloud analytics and cross-site coordination pause; they sync automatically on reconnect. There is no cloud dependency on the critical path.

How are rules defined? Do I need to write firmware?

No firmware. Behaviour is a small declarative YAML DSL — when/then blocks over the normalized signal space. You push rules to one device or the whole fleet over the REST API and they hot-reload without a reboot. See the engine for the full DSL.

What happens when two rules conflict?

The engine resolves deterministically by a fixed three-key sort: explicit priority, then signal freshness, then specificity. Every resolution is written to the audit log with both contenders and the deciding key. It never silently guesses.

Is there really machine learning on the device?

Optionally, yes. The Cortex-M7 runs quantized TFLite Micro models (up to 256 KB) with sub-millisecond inference, fully offline. The model’s output is just another input your rules can reference, e.g. when: ml.anomaly_score > 0.8.

How many devices can I mesh together?

A mesh scales from 3 to roughly 200 devices, gossiping shared state and rule updates with sub-2-second propagation. Beyond that, layer the edge + cloud hybrid and run multiple meshes across sites.

How secure is it?

Firmware updates are signed with A/B partitioning and automatic rollback. The audit log is hash-chained, so a deleted record is detectable. On-device ML means sensor data never has to leave the building. For regulated environments, the log is your evidence trail.

What does it cost?

Pilot pricing depends on volume and topology. We’re onboarding the first fifty teams personally — tell us what you’re building and we’ll quote a pilot, not hand you a price list.

How long does deployment take?

A single hub is a power drop and a config push — minutes. A mesh is the same per node plus auto-discovery. Most pilots are orchestrating something real on day one and tuning rules by week one.

Can I integrate it with systems I already have?

That’s the point of the open REST + WebSocket API and the YAML rule layer. Outputs address actuators by a stable domain.zone.id path, so the Bobulator drives anything from a relay board to a cloud ingress without vendor lock-in.

What’s on the roadmap?

See the changelog for shipped v1.0 features and what’s coming next — multi-site rule templates, a richer ML model zoo, and expanded LoRa mesh routing lead the list.