Open-source alternator regulator for lithium battery systems

The regulator works with any alternator in a 12, 24, 36, or 48 V nominal system, P-type or N-type field excitation — in other words, 99% of them. An internally regulated alternator will first need to be converted to external regulation; that is a separate project, and there are many tutorials for it already.

Yes — while there are ways around this and you're free to do what you want, running two from one is discouraged.

Not for basic regulator operation. Extra features — over-the-air firmware updates, remote support and diagnostics, extended memory, configuration sharing, leaderboards, the solar forecast — do require an internet connection, commonly via the ship's network or a cell-phone hotspot.

Yes, as safe as possible for something charging a battery. If the ESP32 or its firmware fails completely, a pure-hardware overvoltage stage still prevents runaway alternator output. Before that, six software protection groups watch currents, voltages, temperatures, and fault on missing or out-of-range values.

A watchdog supervises the control loop. Every meaningful function is profiled and has visible timing. If the loop stalls, the processor reboots itself, and the cause is reported.

This product replaces the factory regulator and does not require a DC/DC charger. It allows any alternator to charge lithium (or any other chemistry) batteries directly, at programmable rates safe for the whole system.

The two can coexist — a DC/DC charger goes on doing its job for whatever source feeds it. What the regulator removes is the need to push alternator output through a converter rated well below the alternator's capability (typical marine units are 30–50 A) and to convert that energy a second time. Charge rate is then set by what the engine, belt, alternator, and battery can actually take, not by the converter in the middle.

The X Engineering regulator is open source — transparent and freely modifiable. It also has more functionality and costs less (likely due to no office, no dealer chain, no engineering team to pay for). X Engineering is possibly the first case of an AI-enabled company for a hardware product, at least in the marine industry, and it has greatly benefitted from Claude Code and free support / idea generation / peer review on internet forums. The downside? New product, not as proven as the incumbents.

Please do. The platform is built for owners who want to tweak and extend — especially now that AI coding tools make firmware customization realistic for motivated but non-expert owners. One recommendation: it is preferable to submit feature / pull requests first, and only fork the code as a last resort, to try to keep the future manageable for the fleet.

The documentation covers installation and setup step by step. For everything else there is the support forum, or email joe@xengineering.net.

Yes. An engine-speed limit table has ten editable breakpoints (100 to 4600 RPM by default), each one a ceiling in amps — or in kW if you would rather think in belt power. This can be used as a temperature limiter, but the regulator also separately includes closed-loop alternator temperature control. You set a limit and the regulator holds output just under it. It regulates against a 60-second projection of where temperature is heading rather than only where it is, so it eases off before the limit instead of overshooting and recovering. Above the limit sit two more tiers: a warning threshold that ramps the field to zero and starts a lockout, and a critical threshold that cuts the field immediately.

The gain over a fixed field percentage is that you keep full output while the alternator is still cold and give up output only when temperature actually demands it. If you prefer the conservative version, set the table low and let it be the only limit. The automatic temperature control works well where it has been tested, but every installation is a little different.

Yes, there are comprehensive safeties which both warn you (external buzzer, notifications over CAN) and automatically reduce then cut the field to protect equipment.

Technically no — the regulator runs fine without one, but a battery shunt is highly recommended. What it adds is the fastest layer of system protection: it measures current thousands of times per second, so the regulator sees a sudden load disconnection (load dump) as a current step — often a much easier signal to detect than the resulting voltage rise, especially on larger banks.

Example: a 100 Ah bank, alternator delivering 100 A because the microwave and a water heater are running, and both switch off at once. Without a battery shunt, the first evidence is the voltage rising — and by then the full alternator output has already been diverted into a bank that may not want anywhere near that charge rate. With the shunt, the current step itself triggers the response, earlier and smoother. The shunt also enables battery-current-limited charging (holding a set charge current into the bank regardless of what the loads are doing), and gives you a highly accurate 20-bit battery monitor (state of charge) at no extra cost.

The shunt is recommended in all cases, but it is most essential with smaller banks and more delicate alternators.

Yes, for monitoring: the regulator listens for the bank's voltage, current, temperature, and state of charge on the NMEA 2000 network (Battery Status and DC Detailed Status PGNs, 127508/127506, on a user-selectable battery instance) and shows and logs them alongside its own measurements.

Protection and the fast control loop stay on the regulator's own wired sensors, always. Bus data arrives about once per second and lags reality by a second or two — fine for display and for the slow charge-limit coordination described below, far too slow to catch a load dump, a sudden RPM change, or a BMS disconnect. Fast events need millisecond response, and that comes only from directly wired sensors.

The battery shunt is read by a 20-bit precision measurement chip (TI INA228) — the same class of part used in laboratory power analyzers. Its dynamic range resolves milliamps of trickle current while still measuring hundreds of amps, and it samples fast enough to feed the protection layers described above, not just slow state-of-charge accounting. It tracks closely with the popular BMV-712.

The standard alternator-current sensor is a clamp-on Hall-effect sensor (±200 A / ±300 A / ±500 A ranges available; 300 A ships as the default). It installs in seconds — it clamps around the output cable, nothing is disconnected. In principle the regulator's precision shunt input can be repurposed to read an alternator-side shunt instead of a battery shunt, but that is a customization; the supported configuration is the Hall sensor on the alternator and the shunt (if fitted) on the battery. If your installation already has alternator-side shunts, get in touch for further discussion.

Yes. The regulator reads the alternator's stator signal (W-terminal) and converts it to engine RPM using your pulley ratio. That RPM drives the regulator's own RPM-indexed charge tables (the belt-load management), and it can optionally be broadcast onto NMEA 2000 as engine speed (PGN 127488) for your displays. Treat the absolute number as advisory — it is pulley-ratio math on an electrical frequency, not a calibrated tach — but it is exactly what belt-load logic needs. It works even at idle with the field off: all it needs is 25 mV on the low end, and it tolerates up to 100 V on the high end.

The regulator supports daisy-chained digital temperature sensors (OneWire), and by default its own board temperature sensor doubles as a battery-temperature proxy when the regulator is mounted near the batteries — one reason near-the-bank mounting is recommended. Charging is blocked below freezing to protect lithium banks. More convenient wiring for additional temperature sensors is planned for a future version — a convenience improvement, not a functionality difference.

The first three can live on one wire at 250 kbps simultaneously. BMS-Can is the odd one out: different speed, different ID format, physically separate segment, cannot coexist, and is not common on boats, which favor NMEA 2000.

All CAN parameters are used for display and slow decision making. CAN is not fast or reliable enough for critical control loops.

Yes — that is the intended pattern. Set a temperature alarm on the regulator's temperature instance in your MFD. If the regulator's temperature sensor fails or goes stale, the regulator stops transmitting that PGN entirely rather than sending a stale number, which trips the "data lost" alarm every major MFD offers. A silently failed temperature sensor is how alternators cook; this closes that hole with the alarm infrastructure you already own.

Yes, today, with no extra configuration on the regulator: Signal K servers read the NMEA 2000 network, so everything in the table above flows into Signal K automatically, where you can set threshold alarms, push notifications to your phone, and build dashboards. A direct WiFi connection to Signal K (for boats with a Signal K server but no NMEA 2000 backbone) is a possible addition if someone requests it.

Three paths work right now:

• A charge limit on the CAN bus — the protocol table above. A published charge-current limit of zero is obeyed as stop-charging.
• A hardwired contact: the regulator has a dedicated enable/disable digital input, opto-coupled, accepting any signal from 5 to 28 V. Managed BMSes generally have an allow-to-charge output meant for exactly this — REC and Lithionics units ship programmable outputs whose stated purpose is switching off chargers that don't speak their protocol. A Cerbo GX relay can drive the same input, though relay 1 ships programmed for generator start/stop, so using a GX for charge control means the second relay or Node-RED.
• Over the network: the regulator joins the boat's WiFi network, and every control in its web interface is a plain HTTP call, so anything that can make an HTTP request — Node-RED on a Cerbo GX, Home Assistant, a shell script — can turn charging on or off or change settings. The interface is documented and open.

You can. Cloud features are a single on/off switch in the interface. NMEA 2000 transmit ships off until you turn it on, message by message. Anything you don't enable does not run. With everything switched off it is a self-contained local device: web dashboard on the boat's network, no account, no internet.

The extra features exist because the marginal cost of adding them was negligible and the fleet data will allow interesting queries. Some examples: predicted passage and anchorage comfort based on data from similar boats in similar weather. Vessel performance comparisons (best <30-foot VMG upwind in 30 knots of wind with a motion sickness index under X). Alternator and battery lifetimes by manufacturer and use case. What should the temperature limit of an alternator actually be, and how does it affect life? The list goes on once the database is built.

None of the above is allowed near the critical control path: the field control loop runs on its own ~30 ms cadence, and the slow work — writing logs to flash, uploading to the cloud — is deliberately deferred until the alternator field is off.

Since the firmware is open source, deleting what you don't like is always possible, but by switching it off instead you stay on the maintainable path, which is better for most users.

One thing, and only if you switch it on: Defer to Solar, which rests the alternator when the solar forecast says your panels will cover the next few days. It fetches from Open-Meteo, a public weather service, and that request carries your latitude and longitude. It doesn't go through xengineering servers, so the cloud switch doesn't govern it — it's a separate switch, and it ships off. With both off, nothing leaves the boat.

A 0–100 score for how comfortable an anchorage was, computed from the on-board 6-axis motion sensor. It measures how badly the boat rolls and pitches — heel weighted 65%, pitch 35%, full penalty at 12° of heel or 8° of pitch. It activates below 1.3 kt and is grayed out underway. Metrics are subject to change as the fleet gets calibrated.

Underway, the same sensor produces a passage comfort score and a motion sickness index instead.

Near the battery bank is preferred: the board's temperature sensor then serves as a battery-temperature proxy for the freezing lockout and temperature-aware control tuning. Distance from the engine is not a problem — see the next question.

A bulkhead or wall is the preferred location. The enclosure is solid wood, so do not mount it directly to the engine — look at where other wood is already living happily on the boat, and it will be fine in the same sort of place.

Above the waterline, and the higher the better, as with any electronics. The board is conformally coated and will resist splash, and with dielectric grease in the connectors it is about as robust as an electrical product gets short of full sealing.

No. At the recommended 14 AWG, a 6 m run drops about half a volt at worst-case field current, and the regulator simply compensates with a slightly higher field duty cycle. (Fields rarely run near 100% duty in any case — a field driven flat-out overheats the alternator long before wiring drop matters.)

Solid cherry hardwood, in two pieces: a wall dock that mounts to the bulkhead, and a base holding the board that slides down onto it on a 45° sliding dovetail. It is finished with Milesi XGC065 — after trying a few other options, this one has the best moisture resistance.

The enclosure's full CAD (STEP, DXF, drawings) is published in the documentation — print, machine, or build your own housing if you prefer. The box can be omitted from a shipment on request, but the price does not change: one product, one price, kept simple where possible.

No. The board's own consumption, measured at the DC input with the field off, is 0.7 W on a 12 V system and 1.31 W on a 48 V system; in the common idle state it drops to 0.27 W. The field is driven by a MOSFET with 7.2 mΩ on-resistance, so a typical 4–5 A field adds well under a quarter of a watt, and even a 15 A field — far above marine norms, and something the wiring won't handle — adds about 1.6 W spread into the board's copper.

Nothing here is even close to needing a heatsink, a fan, or clearance for airflow.

Within 2 business days of ordering unless otherwise noted.

No — United States addresses only for now. Join the international waitlist and you'll get an email when shipping opens to your country. Questions: joe@xengineering.net.

Officially no, but I will work with you in special cases.

Officially no, but I will be reasonable in the event of failures. Reputation is essential for a new small business.