Skip to content
BranchesTags

Repository files navigation

Solar Home Gateway

A brand-agnostic SunSpec Modbus TCP gateway for the Solplanet ASW H-S2 series. Polls the inverter over HTTP CGI every 5 seconds and re-exposes the readings as a SunSpec-conformant Modbus TCP server, so Home Assistant's built-in SunSpec integration can read it natively — no custom component required.

Why

Most solar inverters ship with vendor-locked APIs. Solplanet is no exception: its data lives behind an undocumented HTTP CGI endpoint. Re-publishing the data over a standard, vendor-neutral protocol (Modbus + SunSpec) decouples the inverter from the home-automation layer, so a future inverter swap is a one-line config change instead of a Home Assistant rewrite.

Table of contents

Architecture

Three layers, decoupled by typed contracts. The Modbus server never touches HTTP, and the adapter never touches Modbus. The state bus is the only object either layer is allowed to touch.

flowchart LR
  subgraph ADAPTER [Adapter — transport only]
    SP[Solplanet CGI<br/>HTTP GET /getdevdata.cgi]
  end
  subgraph DOMAIN [Domain core — typed contract]
    BUS[InverterStateService<br/>double-buffered bus<br/>stale policy]
  end
  subgraph PROTOCOL [Protocol surface — wire only]
    MB[SunSpecModbusServerService<br/>M1 + M101 read-only<br/>IServiceVector]
  end
  subgraph CONSUMER [Consumer]
    HA[Home Assistant<br/>SunSpec integration]
  end
  SP -- poll every 5s --> BUS
  BUS -- refreshFromBus atomic swap --> MB
  MB -- TCP :5020 --> HA
Loading

The type system enforces the seams:

  • InverterAdapter exposes only read(): Promise<InverterState>. The concrete SolplanetCgiAdapter is bound via the INVERTER_ADAPTER DI token ({ provide: INVERTER_ADAPTER, useExisting: SolplanetCgiAdapter }). Swap the concrete class to add a new inverter brand — the rest of the system is unchanged.
  • SunSpecModbusServerService exposes only Modbus handlers and never imports @nestjs/axios. If it ever needs network data, the seam is broken — push the call into a new adapter.

The double-buffer swap

The Modbus server owns two register blocks (bufA, bufB). Each poll tick:

  1. The adapter writes a fresh InverterState into the inactive buffer.
  2. The service atomically flips active from 'A' to 'B' (or vice versa).

Because single-threaded JS makes the assignment truly atomic, a concurrent Modbus read either sees the entire pre-write or the entire post-write block — never a torn mix where, e.g., W is fresh but W_SF is stale. This is critical because SunSpec clients pair a raw value with its scale factor in the same read; a torn pair would silently corrupt the displayed value.

Lifecycle

Phase Hook What happens
Boot OnApplicationBootstrap Server seeds both buffers from bus.snapshot(), then binds the TCP socket
Steady state @Cron(EVERY_5_SECONDS) adapter.read() → bus.publish() → modbus.refreshFromBus()
Shutdown OnApplicationShutdown Server closes its socket within SHUTDOWN_TIMEOUT_MS (5 s)

enableShutdownHooks() in main.ts is what wires SIGTERM/SIGINT into the shutdown hook.

Quick path

nvm use && corepack enable && pnpm install --frozen-lockfile
cp .env.example .env             # then set INVERTER_SN
pnpm run start:dev

The gateway binds Modbus on :5020 and HTTP on :3000. Verify with:

curl http://localhost:3000/healthz          # {"status":"ok"}

Without a real inverter, the cron will warn every 5 s — that's the adapter returning the offline state, not a crash.

Prerequisites

Tool Version Notes
Node.js ≥ 24 (per .nvmrc) LTS — use nvm or fnm
pnpm 11.x (pinned via packageManager) corepack enable activates it
OS macOS / Linux Windows not tested; native build skipped via optional=false

Installation

nvm use                 # honour .nvmrc
corepack enable         # activate pnpm@11 from packageManager field
pnpm install --frozen-lockfile

The optional=false line in .npmrc and the allowBuilds entries in pnpm-workspace.yaml exist only to skip the serialport@13 native build. The gateway is TCP-only, so the native build would just waste time and occasionally fail on macOS. Do not "fix" these without reading this section first.

Configuration

Copy .env.example to .env and edit. Every port is validated at boot — out-of-range → the process crashes with a descriptive log line. This avoids discovering the misconfiguration five minutes later when the Modbus server fails to bind.

Key Default Range Purpose
INVERTER_BASE_URL http://192.168.1.50:8484 http(s)://... CGI base; getdevdata.cgi appended by the adapter
INVERTER_DEVICE_ID 2 — Solplanet device id (query param)
INVERTER_SN "" (required) non-empty Inverter serial (query param) — empty string refuses to boot
INVERTER_TIMEOUT_MS 4000 > 0 CGI wall-clock budget
POLL_INTERVAL_MS 5000 > 0 Sched tick (decorator-fixed at 5 s in v1)
POLL_TIMEOUT_MS 3000 > 0 Adapter-internal abort budget
MODBUS_HOST 0.0.0.0 — Must be a private interface in production
MODBUS_PORT 5020 1..65535 Non-privileged
MODBUS_UNIT_ID 1 1..247 Modbus slave id (reported in M1.DA)
STALE_AFTER_MS 30000 > 0 snapshot() stale threshold
SHUTDOWN_TIMEOUT_MS 5000 > 0 Graceful close budget
HTTP_PORT 3000 1..65535 /healthz only

Security note: Modbus TCP has no authentication and no encryption. Anyone reachable on port 5020 can read (and, on a writeable gateway, set) the inverter state. Run the gateway on a trusted VLAN only; set MODBUS_HOST to a private interface in production (192.168.1.x or similar) and rely on network segmentation. Modbus TLS is out of scope for v1.

Running

pnpm run start:dev     # ts-node, transpile-only (no hot reload)
pnpm run build && pnpm run start   # compiled JS, production-style

Expected log lines on first boot:

[Nest] ... LOG [Bootstrap] HTTP listening on :3000
[Nest] ... LOG [SunSpecModbusServerService] SunSpec Modbus TCP server bound to 0.0.0.0:5020 (unitID=1)

The poller then logs inverter CGI http failed: ... every 5 s if no inverter is reachable — that's normal offline behaviour, not a crash.

Docker

A multi-stage Dockerfile and a docker-compose.yml ship with the repo. The image is ~227 MB and runs as a non-root nestjs user (uid 1001) on node:24-alpine.

Quick path (Docker)

# 1. Copy .env.example → .env and set INVERTER_SN (required).
cp .env.example .env

# 2. Build the image and start the gateway.
docker compose up -d --build gateway

# 3. Verify it's up.
curl http://localhost:3000/healthz          # {"status":"ok"}

# 4. Tail the logs.
docker compose logs -f gateway

The compose file maps:

Host Container Purpose
:5020 5020 Modbus TCP — Home Assistant connects here
:3000 3000 HTTP — /healthz liveness probe

Inside the container the gateway defaults INVERTER_BASE_URL to http://host.docker.internal:8484, so it can reach a Solplanet inverter running on the host. This works out of the box on macOS and Windows Docker Desktop; on Linux either:

  • run docker compose run --network host gateway, or
  • start the IoTech simulator profile and point the gateway at modbus-simulator:5020 instead.

Dockerfile highlights

# Stage 1: builder — node:24-alpine, pnpm@11 from packageManager,
# full install + `pnpm build`.
FROM node:24-alpine AS builder
# ... pnpm install --frozen-lockfile --ignore-scripts
# ... pnpm build
# ... pnpm prune --prod

# Stage 2: production — copy pruned node_modules + dist, non-root.
FROM node:24-alpine AS production
# ... adduser -S -u 1001 nestjs
USER nestjs
EXPOSE 3000 5020
HEALTHCHECK CMD wget --spider http://localhost:3000/healthz || exit 1
CMD ["node", "dist/main.js"]

--ignore-scripts skips husky install (which would fail because the build context has no .git/) and the optional native build of serialport@13 (the gateway is TCP-only and never needs it).

Env vars (Docker)

docker-compose.yml reads INVERTER_SN (and any other secrets) from the host .env file. The compose file overrides a few defaults for containerized use:

Variable Container default Why
INVERTER_BASE_URL http://host.docker.internal:8484 reach a host-attached inverter from inside the container
MODBUS_HOST 0.0.0.0 bind all container interfaces (only mapped host ports are reachable from outside)
MODBUS_PORT 5020 the canonical SunSpec/Modbus port
HTTP_PORT 3000 liveness probe

Everything else comes from .env.

Resource limits

docker-compose.yml caps the gateway at 256 MB RAM and 0.5 CPU — more than enough for the cron + Modbus server. Increase if you run a much larger fleet behind one container.

Tear down

docker compose down              # stop + remove containers + default network
docker compose --profile simulator down   # also drop the simulator
docker image rm sunspec-gateway:local     # free the build cache

IoTech Modbus simulator

A second compose service, modbus-simulator, ships a generic Modbus TCP server based on the IoTech iotechsys/pymodbus-sim:1.0 image. It is NOT a SunSpec simulator — it speaks bare Modbus TCP and is useful only for verifying that some Modbus server is reachable on the network when you don't have the real inverter at hand. The gateway's M1/M101 register map will read as zeros from the simulator.

It is gated behind the --profile simulator flag so docker compose up stays minimal:

# Start only the simulator (host port 5021 → container 5020).
docker compose --profile simulator up -d modbus-simulator

# Smoke-test with pymodbus.
python -m venv .venv && source .venv/bin/activate
pip install pymodbus==3.*
python - <<'PY'
from pymodbus.client import ModbusTcpClient
c = ModbusTcpClient("127.0.0.1", port=5021)
c.connect()
print(c.read_holding_registers(0, 2).registers)   # arbitrary response
PY

If you want the gateway to talk to the simulator instead of a real inverter, set INVERTER_BASE_URL to http://modbus-simulator:5020 on the gateway service and start both with --profile simulator. The simulator's host port (5021) is different from the gateway's (5020) to avoid collisions.

Reference: https://docs.iotechsys.com/edge-xrt21/simulators/modbus/overview.html.

Verifying

1. HTTP liveness

curl http://localhost:3000/healthz
# {"status":"ok"}

/healthz is deliberately cheap — it answers as long as the Nest HTTP server is up. It does not check whether the inverter is reachable; if you need that signal, add a separate /readyz endpoint.

2. Modbus with pymodbus

python -m venv .venv && source .venv/bin/activate
pip install pymodbus==3.*
from pymodbus.client import ModbusTcpClient
from pymodbus.payload import BinaryPayloadDecoder
from pymodbus.constants import Endian

c = ModbusTcpClient("127.0.0.1", port=5020)
c.connect()

# SunS magic at 40000-40001
suns = c.read_holding_registers(0, 2).registers
assert suns == [0x5375, 0x6E53], f"bad SunS magic: {suns!r}"

# M1 ID and L at 40002-40003
m1 = c.read_holding_registers(2, 2).registers
assert m1[0] == 1 and m1[1] == 68

# M101 ID and L at 40070-40071
m101 = c.read_holding_registers(70, 2).registers
assert m101[0] == 101 and m101[1] == 52

# M101.W (AC power, W) at 40084 with W_SF at 40085
w  = c.read_holding_registers(84, 2).registers
raw, sf = w[0], (w[1] - 0x10000) if w[1] > 0x7FFF else w[1]
print(f"AC power: {raw * 10**sf:.1f} W")

# M101.ST (operating state) at 40108
st = c.read_holding_registers(108, 1).registers[0]
# 1=OFF, 2=SLEEPING, 3=STARTING, 4=MPPT, 5=THROTTLED, 6=SHUTTING_DOWN, 7=FAULT, 8=STANDBY
print(f"Operating state: {st}")

Address convention

modbus-serial v8 invokes our handlers with the offset (addr − 40000), not the absolute Modbus address. So register 40084 (M101.W) is read as getHoldingRegister(84). The tables below use absolute Modbus addresses (40000-based); subtract 40000 to get the offset that shows up in handler calls.

Home Assistant integration

Add the SunSpec integration and point it at the gateway's host (modbus: → host: → port 5020, slave 1). HA will discover the gateway automatically by reading the SunS magic at 40000 and walking the model headers.

Register map (verified against sunspec/models)

Address Field Type Notes
40000–40001 SunS magic 0x5375 0x6E53 discovery sentinel
40002 M1.ID uint16 = 1 (Common)
40003 M1.L uint16 = 68 (model length in registers)
40004–40019 M1.Mn string16 vendor name, NUL-padded (32 chars)
40020–40035 M1.Md string16 model name (32 chars)
40036–40043 M1.Opt string8 options, zeroed in v1
40044–40051 M1.Vr string8 version, zeroed in v1
40052–40067 M1.SN string16 serial number (32 chars)
40068 M1.DA uint16 Modbus unit ID (from MODBUS_UNIT_ID)
40069 M1.Pad uint16 canonical even-alignment pad
40070 M101.ID uint16 = 101
40071 M101.L uint16 = 52
40072 M101.A uint16 AC current (A)
40073–40075 M101.AphA/B/C uint16 zeroed (single-phase)
40076 M101.A_SF int16 shared by A / AphA / AphB / AphC
40077–40079 M101.PPVphAB/BC/CA uint16 DC voltage phases (zeroed)
40080 M101.PhVphA uint16 AC voltage phase A (V)
40081–40082 M101.PhVphB/C uint16 zeroed (single-phase)
40083 M101.V_SF int16 shared by PhVph* and PPVph*
40084 M101.W int16 AC power (W) — can be negative for consumption
40085 M101.W_SF int16
40086 M101.Hz uint16 grid frequency (Hz)
40087 M101.Hz_SF int16
40088–40093 (VA / VAr / PF / DCA / DCV / DCW placeholder) zeroed
40094–40095 M101.WH acc32 BE lifetime energy (Wh) — divide by 1000 for kWh
40096 M101.WH_SF int16 typically 0
40097–40107 (DC and temperature block) zeroed in v1
40108 M101.St enum16 operating state (1..8)
40109 (pad)
40110–40121 (event bitfields) zeroed
40122 EOM sentinel 0xFFFF end-of-models marker
40123 reserved zero padding

Every register is big-endian uint16 on the wire. Multi-register values (acc32, strings) split the high word first.

Operating states

M101.St is an enum16 — the numeric values ARE the SunSpec enum, not arbitrary codes.

Value Name Meaning
1 OFF inverter not producing (night, fault, shutdown, stale)
2 SLEEPING low-power standby, not feeding the grid
3 STARTING wake-up / self-test sequence
4 MPPT running and feeding the grid (most common daytime state)
5 THROTTLED producing, but output is capped (grid rule, temperature, …)
6 SHUTTING_DOWN graceful shutdown in progress
7 FAULT error state — see inverter display for details
8 STANDBY ready but not actively producing (e.g. grid absent)

The adapter maps Solplanet's stringy STATE field ("online", "sleeping", "fault", …) onto these values, with the following rules:

  • online / mppt / running / grid-connected / normal → MPPT
  • sleeping / standby / idle → SLEEPING
  • starting / startup / init → STARTING
  • fault / error / alarm → FAULT
  • anything else (including off, shutdown, stopped, empty) → OFF

Scale-factor math

SunSpec represents every measurement as a raw × 10^SF pair, where raw fits in int16 (-32768..32767) and SF is itself an int16. The math lives in src/modbus/scale-factor.ts.

encode(value) returns { value: raw, scaleFactor: sf }:

  • Whole numbers → prefer SF = 0 (no precision loss). If the value overflows at SF = 0, walk positive SFs (downgrade precision) until it fits or we hit SF_CEIL = 10.
  • Fractional values → walk negative SFs (most precision first) until it fits or we hit SF_FLOOR = -10.
  • Overflow at the floor → clamp raw to INT16_MAX.
  • Non-finite (NaN, Infinity) → return { value: 0, scaleFactor: 0 }.

Examples:

Input SF raw Decoded as
0 0 0 0
230.5 (V) -1 2305 2305 × 10⁻¹ = 230.5
4500 (W) 0 4500 4500 × 10⁰ = 4500
50000 (W) 1 5000 5000 × 10¹ = 50000 (downgraded)
1e15 -10 32767 clamped to int16 max

Clients must read value AND scaleFactor in the same Modbus transaction — that's why the double buffer matters. Reading them separately can mix two different poll ticks and produce nonsense.

Stale data handling

InverterStateService (src/state/inverter-state.service.ts) is the single source of truth for "is the data fresh?".

  • On every fresh publish (isStale: false) → stamp lastUpdatedAt = Date.now() and update lastGood.
  • On a stale publish (isStale: true, e.g. adapter hit an HTTP error) → preserve the previous lastUpdatedAt so the 30 s threshold keeps tripping. lastGood is not updated.
  • On snapshot() after STALE_AFTER_MS (default 30 s) without a fresh publish → production fields are zeroed, operatingState = OFF, isStale = true, and lastUpdatedAt = Date.now(). Identity fields (vendorName, modelName, serialNumber) and lifetimeEnergyKwh are preserved from lastGood so the inverter stays identifiable during an outage.

The adapter may pass any lastUpdatedAt (typically 0) — the bus stamps the real one. This contract is what lets every adapter stay trivially simple: it just returns the truth, the bus owns time.

The adapter contract

InverterAdapter (src/state/inverter-state.types.ts) is the abstract class every transport implements:

abstract class InverterAdapter {
  abstract readonly vendorName: string;
  abstract readonly modelName: string;
  abstract read(): Promise<InverterState>; // NEVER throws
}

read() MUST NOT throw. Every failure mode — HTTP timeout, non-2xx, connection refused, malformed JSON, non-numeric values — must resolve to an offline InverterState with isStale: true, operatingState = OFF, zeroed production values, and identity preserved from config.

This contract is what keeps the cron trivial: it can await the adapter without try/catch (one is included as belt-and-braces, but it should never fire).

The Solplanet adapter (src/gateway/inverters/solplanet-cgi.adapter.ts) implements this by:

  • Using firstValueFrom(http.get(...).pipe(timeout(...), catchError(...))) with a wall-clock guard independent of the axios-level timeout, so a stalled TCP socket can't wedge the cron.
  • Accepting both string and numeric STATE values (older firmware emits numeric codes, newer firmware emits strings).
  • Coercing every numeric field via toFiniteNumber() so a stray string or NaN becomes 0 instead of poisoning downstream math.
  • Falling back from WH to E-Total for lifetime energy (firmware versions alternate between the two keys).

Troubleshooting

Symptom Likely cause Fix
Process exits at boot with INVERTER_SN is required INVERTER_SN is empty in .env Set INVERTER_SN= to your inverter's serial number
Process exits with MODBUS_PORT must be an integer in [1, 65535] Out-of-range port in .env Check the port value (no extra chars, no negatives)
port 5020 in use at startup Another process bound 5020 (often a pymodbus server, or a prior gateway instance) lsof -nP -iTCP:5020 -sTCP:LISTEN to find the owner, or set MODBUS_PORT to 5021+
Polling logs inverter CGI http failed: ECONNREFUSED every 5 s Inverter is off / unreachable / wrong INVERTER_BASE_URL Verify the inverter is on and reachable; curl <base>/getdevdata.cgi?device=2&sn=<SN> should return JSON
Polling logs non-JSON or empty body A proxy or captive portal intercepted the request Check the network path; the gateway never follows redirects (maxRedirects: 0)
Home Assistant sees the gateway but sensors stay unknown HA polled during a stale window or before the first publish Wait one poll cycle (5 s); HA will pick up the next snapshot
AC power reads as 0 even when the sun is shining The adapter is in an offline window Check the gateway logs for No fresh data — marking state stale
M101.W reads back the wrong value A Modbus client read W and W_SF in two separate transactions Read both in a single FC3 transaction — see the double-buffer note above
pnpm test:e2e fails with Modbus server did not bind to 127.0.0.1:5021 within 5s Port 5021 is held by another process lsof -nP -iTCP:5021 -sTCP:LISTEN; or set MODBUS_TEST_PORT to a free port

Observability

The gateway uses Nest's default logger. Useful log patterns:

Pattern Meaning
SunSpec Modbus TCP server bound to ... Boot succeeded; socket is listening
inverter CGI http failed: ... One poll tick couldn't reach the inverter — not fatal, will retry in 5 s
No fresh data — marking state stale The 30 s stale threshold tripped; production values are zeroed on the next snapshot
Modbus server closed cleanly Graceful shutdown completed within the budget
Modbus server did not close within 5000ms — forcing resolve A client held the connection open past the shutdown budget; the process still exits
Modbus server close error: ... The underlying socket rejected the close — usually harmless, the process still exits

Log lines are intentionally at warn level for inverter failures (debug would be too quiet in a 24/7 deployment; error would drown out real issues).

Security checklist

  • MODBUS_HOST is set to a private interface (not 0.0.0.0) in prod.
  • The host running the gateway is on a trusted VLAN.
  • Firewall blocks external traffic to 5020 and 3000.
  • /healthz is unauthenticated on purpose (liveness probes) — do not expose it to the public internet.
  • INVERTER_SN is not committed to .env (which is .gitignore'd).

Development workflow

pnpm lint          # ESLint v9 flat config + typescript-eslint
pnpm build         # tsc -p tsconfig.json (strict mode)
pnpm test          # Jest unit suite (fake timers for the stale policy)
pnpm test:cov      # Unit + coverage (CI uses this)
pnpm test:e2e      # Boots the gateway on port 5021, drives it via modbus-serial
pnpm run start:dev # ts-node transpile-only (no hot reload)

CI runs lint → build → test --coverage → test:e2e on every PR via .github/workflows/ci.yml (Node 24 from .nvmrc, pnpm 11 from packageManager). The Setup pnpm step must run before setup-node@v6 because cache: 'pnpm' shells out to pnpm to compute the cache key.

Running a single test

# one file
pnpm test -- inverter-state.service

# one test by name pattern
pnpm test -t "marks all production fields stale"

# e2e single file
pnpm test:e2e -- sunspec-modbus

Commit conventions (enforced by husky → commitlint)

  • Header ≤ 72 chars; body/footer ≤ 100 chars/line
  • Lowercase type (feat:, fix:, chore: — never Feat:)
  • Lowercase first letter of subject (feat: add foo, not feat: Add foo)
  • Conventional Commits via @commitlint/config-conventional
  • No Co-Authored-By or AI attribution — use conventional commits only

Agent-specific field guide

AGENTS.md is the companion file for OpenCode / Claude sessions. It covers the compiler-enforced seams, the modbus-serial v8 quirks (sync getMultipleHoldingRegisters return, missing readDeviceIdentification in the type), the OpenSpec change directory, and the local skill registry at .atl/skill-registry.md.

Adding a new inverter brand

The architecture is built for this. Three steps:

  1. Implement the adapter. Subclass InverterAdapter in src/gateway/inverters/<brand>.adapter.ts. read() must never throw — resolve failures to { isStale: true, operatingState: OFF, … }. Set vendorName and modelName for SunSpec M1.Mn / M1.Md.
  2. Wire the adapter. In src/gateway/gateway.module.ts, replace { provide: INVERTER_ADAPTER, useExisting: SolplanetCgiAdapter } with your new adapter. Drop the HttpModule import if your transport doesn't use HTTP.
  3. Extend the env loader. Add brand-specific keys to src/config/configuration.ts. Reuse the existing validatePort / pickNumber / pickString helpers — out-of-range values should crash boot, not the cron five minutes later.

The Modbus server, state bus, and Home Assistant integration are unchanged. Tests in test/e2e/sunspec-modbus.e2e-spec.ts show the FakeAdapter pattern for writing a new E2E without the real hardware.

Project layout

src/
├── main.ts                                # bootstrap (enableShutdownHooks)
├── app.module.ts                          # ConfigModule (global) + GatewayModule
├── config/
│   └── configuration.ts                   # typed env loader + boot-time validator
├── gateway/
│   ├── gateway.module.ts                  # DI wiring for all three layers
│   ├── inverter-polling.service.ts        # @Cron(EVERY_5_SECONDS)
│   └── inverters/
│       ├── inverter.adapter.ts            # re-export (seam convenience)
│       └── solplanet-cgi.adapter.ts       # @nestjs/axios → InverterState
├── state/
│   ├── inverter-state.types.ts            # InverterState, M101_ST, INVERTER_ADAPTER
│   └── inverter-state.service.ts          # double-buffered bus + stale policy
├── modbus/
│   ├── sunspec-modbus-server.service.ts   # IServiceVector + ServerTCP
│   ├── sunspec-registers.ts               # M1/M101 offsets + write helpers
│   └── scale-factor.ts                    # chooseScaleFactor / encode / splitAcc32
└── health/
    └── health.controller.ts               # GET /healthz

test/
├── unit/
│   ├── inverter-state.service.spec.ts     # stale policy, lastGood, timestamp semantics
│   └── scale-factor.spec.ts               # SF math, acc32 split, overflow clamping
└── e2e/
    └── sunspec-modbus.e2e-spec.ts         # boots gateway on :5021, drives it via modbus-serial client

License

MIT — see LICENSE.

References

See also

About

No description, website, or topics provided.

Resources

Readme
MIT license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages