# GREE Controller Self-hosted controller for GREE-compatible air conditioners with a local Web UI, thermostat zones, schedules, Home Assistant integration, history, notifications and a documented HTTP/WebSocket API. **Current release: 0.11.6** > [Full API reference](docs/API.md) — authentication, every endpoint, request bodies, response models, WebSocket events and examples. ## What it does - Discovers and binds GREE Wi-Fi units on the LAN, including mixed protocol generations. - Provides direct technical device control and a separate thermostat layer built around zones. - Supports cooling/heating profiles: Comfort, Sleep, Away and custom quick targets. - Coordinates house, group and per-zone control without silently fighting physical/manual overrides. - Supports schedules, automations, night mode and temporary Quick Thermostat sessions. - Can use GREE sensors, Home Assistant room sensors or a weighted combination. - Stores recent history in SQLite and can archive/query long-term history in InfluxDB 1.x or 2.x. - Exposes native Home Assistant entities through the bundled custom integration. - Includes PWA-capable responsive UI, Polish/English localization and live diagnostics. - Can run in Simulation mode without physical air conditioners. A large warning is shown whenever simulation is enabled. ## UI overview The interface is intentionally split by responsibility: | Area | Purpose | | --- | --- | | Dashboard | Whole-house summary, Quick Thermostat and full manual GREE unit control. | | Devices | Technical unit configuration, binding, connection diagnostics and live ping. | | Zones | Thermostat configuration, profiles, sensors and control ownership. | | Groups | Climate group membership plus optional group-level mode/preset/custom-temperature control. Group ON activates member thermostat control; Group OFF powers member units down and releases group ownership so individual thermostats are not left group-blocked. | | Flow | Full-screen visual logic editor. Flow is the source of truth and is automatically compiled into generated schedules and automations. | | Schedules | Weekly thermostat schedules and ready-made templates, including read-only entries generated by Flow. | | Automations | Time/temperature/Flow-triggered device, group or thermostat actions. | | History | Overview, zones, devices and HA sensors with precise chart tooltips and zoom. | | Simulation | Simulated units for safe testing. | | Night mode | Quiet/Sleep/fan limits during an overnight window. | | Home Assistant | Sensor connection, aliases and restricted integration tokens. | | Settings → Application | Simulation, notifications, metrics/logs, InfluxDB, debug and backup. | | Settings → GREE | Controller identity, polling/discovery, GREE commands and live GREE traffic. | | Events | Operational event log and diagnostics. | The top bar uses the same compact language and theme controls on desktop and mobile. The live debug overlay can be filtered to **All**, **Requests** or **GREE** frames. ## Quick start for development Requirements are installed automatically on supported Debian/Ubuntu environments unless `--no-install` is used. ```bash cp .env.example .env ./scripts/dev.sh ``` Useful development commands: ```bash ./scripts/dev.sh --check # format, tests, build and HTTP smoke test ./scripts/dev.sh --release # optimized local build ./scripts/dev.sh --reset # recreate the local SQLite database ./scripts/smoke.sh # API smoke test against an isolated simulated instance ``` The default panel is available at `http://127.0.0.1:8787` when the process is bound locally. ## LXC / systemd installation The production scripts target Debian/Ubuntu systemd hosts and LXC containers. ### First installation ```bash sudo ./scripts/install.sh ``` Optional flags: ```bash sudo ./scripts/install.sh --skip-tests sudo ./scripts/install.sh --no-start ``` Installed layout: | Path | Purpose | | --- | --- | | `/opt/gree-controller/gree-controller` | Release binary. | | `/etc/gree-controller.env` | Environment configuration; mode `0600`. | | `/var/lib/gree-controller/gree-controller.db` | SQLite database. | | `/etc/systemd/system/gree-controller.service` | systemd unit. | | `/var/backups/gree-controller/` | Update backups used for rollback. | ### Update Run from the extracted source tree of the new release: ```bash sudo ./scripts/update.sh ``` The updater builds/tests first, stops the old service, backs up the binary/unit/environment/database, installs the release, checks `/api/health` and rolls back automatically if startup fails. ### Service helper ```bash ./scripts/service.sh status ./scripts/service.sh health ./scripts/service.sh logs sudo ./scripts/service.sh restart ``` `./scripts/install-lxc.sh` remains as a compatibility alias to `install.sh`. ## Configuration Copy `.env.example` for the full supported example. Command-line arguments exist for the core `Config` options, while most day-to-day runtime settings are also editable in the Web UI and persisted in SQLite. Environment values explicitly supplied for supported runtime overrides win over the persisted value after restart. ### Core variables | Variable | Default | Description | | --- | --- | --- | | `GREE_CONTROLLER_BIND` | `0.0.0.0:8787` | HTTP/WebSocket listen address. | | `GREE_CONTROLLER_DATABASE` | `./data/gree-controller.db` | SQLite database path. | | `GREE_CONTROLLER_APP_TOKEN` | empty | Optional administrator API/Web UI token. Empty means trusted-LAN mode. | | `GREE_CONTROLLER_BASE_PATH` | empty | Optional reverse-proxy prefix such as `/gree`. | | `GREE_CONTROLLER_ID` | `gree-controller` | GREE client/controller identifier. | | `GREE_CONTROLLER_SIMULATE` | `false` | Initial Simulation mode. | | `GREE_CONTROLLER_AUTO_SEED` | `false` | Seed simulated sample data when applicable. | | `GREE_CONTROLLER_POLL_INTERVAL_SECONDS` | `15` | Device polling interval. | | `GREE_CONTROLLER_ZONE_INTERVAL_SECONDS` | `5` | Thermostat control interval. | | `GREE_CONTROLLER_DISCOVERY_TIMEOUT_MS` | `3000` | Discovery timeout. | | `GREE_CONTROLLER_DISCOVERY_BROADCAST` | `255.255.255.255:7000` | GREE discovery target. | | `GREE_CONTROLLER_GREE_INTERFACE` | auto | Optional interface name or local IP used for GREE UDP traffic. | | `GREE_CONTROLLER_COMPRESSOR_PROTECTION_ENABLED` | `true` | Enable controller-side compressor restart/mode-change protection. | | `GREE_CONTROLLER_COMPRESSOR_PROTECTION_SECONDS` | `180` | Compressor protection window (30–1800 seconds; 180 seconds recommended). | Additional environment variables cover history retention, debug, night mode, Home Assistant and InfluxDB. See [`.env.example`](.env.example). ## Connecting physical GREE units 1. Place the controller host on a network that can reach the air-conditioner Wi-Fi modules by UDP. 2. Open **Devices** and start discovery. 3. Auto protocol mode is recommended; the controller accepts both supported GREE encryption generations. 4. Newly discovered devices are bound automatically when possible. Manual bind is also available. 5. Create a thermostat zone for each unit you want the thermostat engine to own. Discovery uses UDP broadcast, so routed/VLAN networks must explicitly permit or relay the required traffic. ### Multi-NIC / dedicated GREE interface For a host with separate management and GREE networks, set for example: ```bash GREE_CONTROLLER_GREE_INTERFACE=eth1 ``` or a local address: ```bash GREE_CONTROLLER_GREE_INTERFACE=192.168.50.2 ``` The controller will use that interface for GREE UDP traffic while keeping the HTTP UI on `GREE_CONTROLLER_BIND`. Diagnostics: ```bash sudo ./scripts/configure-gree-network.sh ./scripts/network-debug.sh ``` The hardened systemd unit allows `AF_NETLINK`, which is required for interface discovery on multi-NIC Linux/LXC systems. ## Thermostat model ### Devices vs zones A **Device** is the physical GREE unit. Device commands are direct/technical commands. A **Zone** is the thermostat owner for a device. It decides demand from room temperature, setpoint, hysteresis, schedules, house rules, manual ownership and safety lockouts. Group control can temporarily apply shared settings. Turning a group OFF powers its member units down and releases group ownership; each thermostat can then be switched back on independently without an OFF group acting as a membership block. After handing an OFF thermostat back to automation, an implicit Comfort target alone does not start it: the zone waits for an active schedule, manual preset/setpoint, Temporary Quick Thermostat, active group/local control or another explicit automatic power intent. ### Compressor protection Controller-side compressor protection is enabled by default and uses a 3-minute window. The switch and duration are available under **Settings → GREE**. Thermostat, group and whole-house starts or Heat/Cool reversals that fall inside the window are kept as visible pending tasks instead of being discarded. The pending task is shown on both the thermostat and its device card. A permanent **Queue** button above Thermostats opens a global modal sorted by execution time, with device/zone, command, owner, protection deadline, live countdown, per-task cancellation and **Cancel all**. Safety OFF commands remain immediate. This distinction is important: direct device control intentionally behaves differently from thermostat control. ### Control ownership The zone state exposes who currently owns control (`control_owner`, `control_source`, timestamps and reason). Physical remote/direct device actions can create a manual-device takeover so normal schedules do not immediately overwrite the user. Explicit zone thermostat control hands ownership back to the thermostat engine when appropriate. Global **Turn all off** is an immediate bulk thermostat action: it powers every enabled unit off and leaves every thermostat locally OFF with no hand-back timer, but it does not create a persistent blocking gate. A later **Resume automation** releases only that zone to automatic arbitration; when no schedule/override/temporary or other explicit intent is active, it stays OFF. **Turn all on** is different: it is an explicit whole-house ON intent, releases local OFF markers and sends ON to enabled units; compressor protection may queue protected starts. Explicit local/group/manual control still has priority. ### Profiles and schedules Each zone stores separate cooling and heating temperatures for: - Comfort - Sleep - Away - Custom/manual target Schedules select profiles or a custom target for selected ISO weekdays (`1=Monday`, `7=Sunday`). Overnight windows are supported. Overlapping enabled schedules for the same zone are rejected. ### Visual Flow **Flow** is the authoring layer for combined schedules and automations. The full-screen editor stores a block graph as the source of truth and compiles it automatically: a pure weekday + time-range + thermostat path becomes a native schedule, while richer logic becomes a generated automation. Generated schedule/automation rows carry their Flow provenance and are read-only in the legacy editors. Stateful blocks include optional 5-field CRON triggers, continuous `stable_for`, bounded state duration, change/edge detection, rolling execution rate limits, delay-before-action, rolling mean/median windows and oscillation detection; their runtime state survives restarts. Generic Home Assistant service actions can call any configured `domain.service` with an optional entity and JSON service data. The reusable Flow preset library is stored as standalone JSON files in `presets/`. Each file contains its name, description and complete Flow graph, while the UI loads the embedded preset catalog dynamically instead of hard-coding graphs in JavaScript. The library has category tabs, search, local favorites/recent presets, a graph preview and requirement checks before applying a preset. Available blocks include weekday/time/date ranges, the application Night mode, outdoor/device/zone temperature, house/device/zone/group state, arbitrary Home Assistant state/numeric/attribute/availability checks, a constant diagnostic source, AND/OR/NOT, thermostat actions, direct GREE actions and group actions. The GREE action uses the existing `DeviceCommand` fields (power, mode, target, fan, swing, quiet, turbo, light, air, xfan, health and sleep). Multiple branches can feed an action; the compiled runtime preserves the graph logic. Quick preset/setpoint overrides normally hand control back at the next schedule boundary. Temporary Quick Thermostat adds explicit start/finish rules such as duration, exact time, temperature reached/stable or next schedule boundary. Implementation and concurrency invariants are documented in [`docs/FLOW.md`](docs/FLOW.md). ## Home Assistant There are two independent Home Assistant directions. ### 1. Home Assistant as a temperature source Configure **Home Assistant / Sensors** in the Web UI or use: ```text HA_URL= HA_TOKEN= HA_ENTITY_ID= HA_OUTDOOR_ENTITY_ID= HA_SENSOR_STALE_AFTER_SECONDS=300 HA_ALLOW_INVALID_TLS=false ``` Per zone, `sensor_source` can be: - `device` — GREE indoor sensor, - `home_assistant` — configured HA room sensor, - `combined` — weighted GREE + HA value. Stale/unavailable external data falls back to the GREE sensor when possible. The optional outdoor sensor is only an assist signal; it never replaces room temperature. ### 2. GREE Controller entities inside Home Assistant Bundled integration directory: ```text home-assistant/custom_components/gree_controller/ ``` Copy `gree_controller` to Home Assistant's `custom_components` directory, restart Home Assistant and add **GREE Controller** from Integrations. Create a restricted token in **Home Assistant / Sensors → Access tokens**. The token secret is displayed only once and the database stores only its SHA-256 hash. The integration uses a restricted `/api/integrations/home-assistant/*` surface rather than administrator endpoints. The integration exposes physical device climate controls, thermostat zone climate entities, whole-house controls, groups and optional unit capabilities. Prefer zone thermostat entities for normal comfort control; physical device entities are direct/manual controls. For entity-ID migration tooling: ```bash python3 scripts/generate_ha_migration.py --help ``` The generated mapping example is under `home-assistant/generated/`. ## History and InfluxDB SQLite is the local hot store. History compaction reduces older local sample density while preserving the resolutions needed by charts. Retention is configurable from the UI or environment. Optional InfluxDB writes samples in parallel and can serve the older part of long history queries: - InfluxDB 1.x: URL, database, optional username/password. - InfluxDB 2.x: URL, organization, bucket and token. If an Influx query fails, the API returns available SQLite data and can include a `storage_warning` rather than discarding the whole history response. See [`GET /api/history`](docs/API.md#history-and-readings) for scopes and bucket behavior. ## Notifications Optional notifications support: - Pushover - Slack webhook - Discord webhook Modes can report problems only or problems plus important state changes. Individual categories include sensor freshness/errors, communication failures, target timeout, automation/control errors and other warnings. Cooldown and failure thresholds are configurable. ## Debugging Enable **Settings → Application → On-screen debug**. The overlay has three filters: - **All** — request diagnostics, application events and GREE frames, - **Requests** — HTTP method/path/status/duration, - **GREE** — sanitized protocol frames when GREE frame logging is enabled. Binding/encryption keys are not intentionally exposed in GREE debug frames. For Linux/network issues also use: ```bash ./scripts/network-debug.sh ./scripts/service.sh logs ``` ## Backup and restore **Settings → Application → Configuration backup** exports settings, devices, zones, groups, schedules and automations. The export intentionally does **not** contain metric history, event history or generated API-token records. It **does contain** GREE binding keys and configured integration secrets, so store it like a credential file. Import replaces application configuration while preserving local metric/event history and generated access-token records. Runtime ownership/timers are sanitized and live device state is re-polled before automatic control resumes. ## Reverse proxy and sub-path deployment For a root deployment, proxy HTTP and WebSocket traffic to the controller normally. For a sub-path, configure: ```bash GREE_CONTROLLER_BASE_PATH=/gree ``` Then serve the application under `/gree/`. The proxy must preserve WebSocket upgrade headers. A typical nginx location is: ```nginx location /gree/ { proxy_pass http://127.0.0.1:8787/gree/; proxy_http_version 1.1; proxy_set_header Upgrade $http_upgrade; proxy_set_header Connection "upgrade"; proxy_set_header Host $host; } ``` If your proxy strips the prefix, keep its upstream/base-path handling consistent. Do not publish the controller directly to an untrusted network without authentication and TLS at the proxy/VPN layer. ## Security The default empty `GREE_CONTROLLER_APP_TOKEN` is intended for a trusted LAN only. Set a long random token when the interface is reachable by other networks. Implemented safeguards include: - optional bearer administrator authentication, - separate restricted generated tokens for Home Assistant, - hashed generated-token storage, - secrets omitted from normal settings responses, - restrictive security headers and no permissive CORS, - API responses marked `no-store`, - hardened systemd service, - safe device power-off behavior when thermostat ownership is removed, - startup synchronization before the thermostat engine begins issuing commands. For remote access, prefer a VPN or authenticated TLS reverse proxy. Treat configuration exports and `/etc/gree-controller.env` as secrets. ## Localization Language packs are JSON files under `lang/`. The build validates/embeds them and generates the public language catalog. To add a language: 1. Copy `lang/en.json` to `lang/.json`. 2. Update `meta.code`, names and locale. 3. Keep the same translation keys as the reference pack. 4. Rebuild the Rust binary. Runtime assets are available at `/lang/index.json` and `/lang/.json`. ## API The HTTP API is the same backend used by the Web UI. It includes devices, zones, groups, house control, schedules, automations, history, control plan, events, settings, backup/restore, debug, tokens and integration tests. Start here: **[docs/API.md — complete API reference](docs/API.md)** Public health check: ```bash curl http://127.0.0.1:8787/api/health ``` Authenticated example when an app token is configured: ```bash curl -H 'Authorization: Bearer YOUR_TOKEN' \ http://127.0.0.1:8787/api/system/info ``` ## Project layout ```text src/ Rust backend, GREE protocol and thermostat engine web/ Embedded Web UI / PWA web/js/ Frontend JS source modules bundled by build.rs lang/ Runtime language packs home-assistant/ Home Assistant custom integration and migration output scripts/ Development, install, update, service and diagnostics systemd/ Production service unit docs/API.md Complete API documentation .env.example Environment reference ``` The application is intentionally self-contained: static assets and language packs are embedded into the binary at build time; SQLite is the default data store; no external frontend build chain is required. `build.rs` concatenates the ordered files from `web/js/` into one generated application bundle and exposes it under a content-hashed URL. ## Validation before release Recommended full check on a host with Rust installed: ```bash ./scripts/dev.sh --check ``` This runs formatting checks, Rust tests, a build and an isolated simulated API smoke test. ## License See [LICENSE](LICENSE). #### Flow diagnostics, portability and concurrency Flow reuses the existing thermostat, group and device-control domains instead of maintaining a second HVAC controller. Thermostat actions update the persistent zone intent and wake the normal zone-control engine, so hysteresis, compressor protection, schedule hand-back, Temporary Quick Thermostat and ownership rules remain authoritative. Flow group actions call the existing group controller; direct device actions use the existing automatic-device action path and `DeviceCommand`. A Flow thermostat OFF also performs the existing forced physical OFF transition. Durable thermostat/device transitions are serialized with the thermostat cycle and then use the canonical zone -> device lock order. The editor supports optimistic revisions, so saving an older copy from another browser/tab returns HTTP 409 instead of overwriting a newer graph. Flow compilation/replacement is serialized with configuration, automation, schedule and thermostat-cycle locks, and generated schedules/automations plus the Flow source are replaced in one SQLite transaction. Runtime condition evaluation happens before the automation execution lock; after acquiring it the rule is reloaded and stale snapshots are discarded. Same-cycle actions claim their target devices deterministically, and Flow/device thermostat writes take the schedule -> thermostat-cycle -> zone -> device path. Home Assistant read failures fail closed instead of making NOT/neq logic accidentally true. Diagnostics include a non-mutating dry-run endpoint/UI with a selectable simulation time, optional per-block sensor/state overrides, a per-node condition trace and an ownership/block reason. Shared Home Assistant Flow inputs can also be tested directly in their editor against the current HA entity state. Flow-scoped execution and dry-run events can be viewed from the editor. Individual Flows can be exported/imported as versioned `gree-controller-flow` JSON. Generated schedules and automations are intentionally excluded from the portable document and are regenerated from the source graph on import. Incomplete Flows can also be saved as disabled drafts; drafts preserve the editor graph but intentionally generate no schedules or automations until completed and saved normally. Shared inputs show which Flows reference them and link directly to those editors. The preset library includes 37 categorized scenarios for comfort, energy, safety, night, reliability, Home Assistant heat-source coordination and advanced multi-branch logic.