Features
Not a menu dump. Capabilities are hung on what keeps you busy at night — aligned with the current implementation.
Features exist to hold three yardsticks: first-pass · remote clarity · data you dare use, and four field stories: plant · campus ledger · OEM · integrator less on-site.
See What we solve and Scenarios.
How to read this page
| You want… | Read |
|---|---|
| How AI cuts debug and support calls | AI: understand · answer |
| Less point matching, no core forks | Less debug: connect |
| Duty can see and act | To the field: see · act |
| Operate after connect | Ops: ledger · care · consumables |
| Trust reports; integrate without lock-in | Data you dare use · less lock-in |
| How private deploy starts | How the base holds (link architecture) |
AI: understand · answer
Lingdian AIOT embeds AI in delivery and triage — not a bolted-on chat box. Tagline: less debug with AI; AI understands and answers; users just use.
| Capability | How you use it | Cost it cuts |
|---|---|---|
| AI onboard | Upload point tables / protocol docs / sample payloads → OCR & recognition → thing-model draft and parse rules → human review then one-click publish | Triple point-matching, night of register copy |
| Lingdian Assistant | Admin float or full-page chat with device / point / alarm / log context; Skills and platform tools to query and propose actions | Repeat support calls; experts guessing remotely |
| Knowledge · Flow · Skill | AI Center assets; complex jobs (e.g. thing-model recognition) as Skills the assistant invokes | Rewriting scripts and scripts-as-docs per project |
| AI-assisted SCADA | Optional layout / binding hints on the diagram side | Dragging from a blank canvas forever |
Principles (same as architecture):
- Humans review: AI drafts; publish / control and other high-risk writes need confirmation.
- Same permissions: assistant = current user; same tenant and RBAC.
- Does not replace the access base: protocol gaps still use plugins; AI speeds “recognize semantics,” not hot-plug codec.
See Design Philosophy. Acceptance: can an unfamiliar table reach a reviewable model? Does asking about this device and this log get a precise answer?
Less debug: connect
| Capability | What you can do |
|---|---|
| MQTT direct / gateway + children | Unified Topic via gmqtt |
| Modbus TCP / RTU | Transparent gateway or master policy with models |
| HTTP · TCP · UDP | Network module + plugins |
| Protocol repo · hot-plug | Publish/activate packs; WASM / Go Plugin |
| Product / device / thing model | Types, property/event/service, pre-register |
| Connect & debug | Params, simulate uplink, downlink; raw + normalized logs |
Point-table recognition: AI onboard above. Details: Device Access chapters. Go is Broker Client only — Architecture.
To the field: see · act
See: dashboards/trends/aggregates · industrial SCADA · mini program (same control-plane APIs). SCADA AI hints: AI: understand · answer.
Act: rule engine & flows · linkage · alarm center · notify center · OTA. Triage Q&A and Skills: AI: understand · answer.
Ops: after connect
Equipment ledger · care plans/tasks · consumables SKU/BOM/stock · region tree · phone jobs (scan, close loops). Maps to the campus ledger story.
Data you dare use · less lock-in
Standard parameters (tree, alias, match, inspect) · dual-track store · data forward (MQTT/HTTP to MES/cloud) · Open API · tenant/project/RBAC/workbench.
Without a data team: standard parameters + dual-track + thing-model-only business hold trust.
Which pack for which story
| Story | Typical packs |
|---|---|
| Plant | Access + models + dashboards/SCADA + rules/alarms + forward |
| Campus ledger | Ledger/care/consumables + notify + mini program |
| OEM | Protocol repo + product templates + tenant/private + OTA |
| Integrator less on-site | Dual-track logs + AI Q&A + remote debug + reusable projects |
How the base holds
Device → Go → Node → Vue (no skip) · gmqtt + LingdianDeviceMessage · PG / TDengine / Redis · start on 4C2G · scale with same boundaries · capacity in Performance. Extend via plugins, Cool modules, Skills, forward scripts, Open API — add modules, don’t fork the core.
Acceptance questions (not menu counts)
- Can an unfamiliar point table reach a reviewable thing model?
- Without a laptop on site, can remote explain last raw + normalized result?
- Can synonym / dirty data be caught or blocked?
- Can a “low supply” style rule go from alarm to action?
- Can one ledger/care step finish on the phone?
If yes, the feature reached the field.