Glossary
Platform Terms
| Term | Description |
|---|---|
| PlantPulse AI | The industrial AI copilot platform as a whole |
| PlantPulse IIoT | The PlantPulse industrial IoT foundation platform (data collection/management) |
| AURA | PlantPulse's in-house industry-specific LLM |
| MCP | Model Context Protocol — the AI-to-data standard integration protocol proposed by Anthropic |
| Integrated MCP | The MCP endpoint provided by the PlantPulse Platform server-web (/api/v5/mcp) — combines factory data and ontology tools |
| Built-in agent | The agent loop embedded in AI Chat Web (AgentOrchestrator + ToolRegistry) — automatically calls tools from Ask. This is distinct from the Automation (formerly "Agent") that users create |
| Feature store | A system that stores and serves precomputed equipment features and LLM context |
| Digital Thread | A service that links factory objects — equipment, work orders, parts, people — and the events that occurred on them in chronological and relational order (plantpulse-digital-thread-api, included by default from 2026.10) → Digital Thread |
| Connector | An MCP endpoint exposed by an external system (MES, SCM, ERP, PLM, QM, etc.) according to a defined spec (mes-mcp, erp-mcp, plm-mcp, qm-mcp). Registered in the Integration Map of the AI admin console |
| Family | The connector type — mes, scm, sched, erp. Each family has a required set of tools, and when a site has multiple connectors of the same family, one of them is the primary |
| Conformance test | A test that verifies a connector responds per spec. It must pass before the connector can be enabled, and must be re-run whenever the endpoint or credentials change |
| Binding | A rule that maps the fields (external IDs) returned by connector tools to object types in the factory graph (equipment, work order, part, …). Bindings may be declared by the connector or drawn by an operator on screen; the operator's binding wins |
| Graph mirror | The factory graph that Digital Thread builds in a dedicated Neo4j by collecting references only from the platform ledger, events, and connectors. It can be rebuilt from the sources at any time |
| Link template (external links) | A per-connector template for the source system's screen address ({base} · {key} · {key.N}). If present, thread nodes get a button such as "Open in SAP"; if absent, there is no link → External links |
| Current values in the source system | The key fields the Digital Thread shows when you click a thread node in the evidence panel, read as that one record from its system-of-record connector. Not stored. «Unknown» if there is no connection → Answer evidence |
| MCP Gateway | A single address in front of every MCP server in the installation (platform integration MCP · AI built-in servers · external servers). It gathers the tools, turns them on and off, shows each caller only its allowed tools, and picks the matching tools for each question → MCP Gateway |
| Tool search terms | Words that link tools with English descriptions to Korean questions (used only in keyword search). One of server declaration · gateway default · operator override is used. Per tool, 8 terms · 64 characters across all languages |
| Automation Sandbox | The service that runs the Python code of automation steps in an isolated process per run. The container itself runs on gVisor, and there is no network inside a run → Automation Sandbox |
Copilot Screen Terms
| Term | Description |
|---|---|
| Copilot | The app people actually use. It finds factory problems and proposes actions with supporting evidence, closes them once a person approves, asks when it doesn't know, and hands repetitive work to Automation — a field decision app. It is used through three things: Ask, Proposals, and Automation → Copilot Screen Map |
| Sidebar | The menu column on the left. From the top: New Chat, Proposals, Automation, then recent conversations, with the profile row at the bottom. There is no separate "Ask" row |
| New Chat | The button at the top of the sidebar. Opens a fresh Ask start screen — this is the way into Ask |
| Ask | The screen where you ask in text, images, voice, or files and get answers with supporting evidence (/chat). It appears first when you open Copilot, and you enter it via New Chat in the sidebar |
| Add file (attachment) | Putting a document or drawing file (PDF · Office · CSV · DXF · scanned TIFF), or a video · audio file, into the Ask input box and asking about that file. The server reads it first (reading · transcription for video and audio), and the AI picks the pages · segments it needs. Only the owner of that conversation sees it, and it is deleted along with the conversation. It is separate from company document search (RAG) → AI Chat — Adding files |
| Today's Tasks | The task summary collapsed into the single line "Today's Tasks: N" on the Ask start screen. Tapping it expands up to five rows in order of urgency. If there are no tasks, the row is not drawn |
| Right panel | A side area that opens while you keep a conversation or proposal detail in place. Source documents, drawings, and video all open here; opening another item on top of the panel adds a Back control to the header |
| Drawing viewer | A new-tab screen opened from the drawing panel via Open in viewer (new tab). Zoom, page list, revision comparison (side by side / overlay), and Ask about this area → Asking with Drawings |
| Proposals | The screen where people decide on and close out the work AI has found (/tasks). A single card flows through detection/intake → approval → action → outcome → recurrence verification, all within the same detail view → Proposals |
| Planned / Unplanned | The two sources of proposals. Planned covers action proposals (daily tasks AI raises in advance, and cards raised by Automation); Unplanned covers anomalies to check and incidents |
| Action proposal | A proposal card on the Planned side — inspection or improvement work that AI raises in advance based on measured data |
| Anomaly to check | An anomaly caught by AI anomaly diagnosis and awaiting human confirmation. It arrives under Unplanned in Proposals, and is either escalated via "Raise as incident" or dismissed with a selected reason |
| Incident | An item raised by a person or escalated from an anomaly to check, an alarm, or a failure precursor — it covers not just failures but every abnormal signal that needs attention. It is closed in the order diagnosis → action plan approval → action execution → outcome recording → verify and close. In code, APIs, config keys, and the DB the name is incident. On screens prior to 2026-10-14 it appears as "Failure Intake" |
| Recurrence verification | The stage that watches, for a defined period after the outcome is recorded, whether the same problem reappears. If it does, the card returns as "Recurrence" |
| Approval draft | A draft created in another system (an MES maintenance request, an SCM order expedite) once a person enters a reason and approves. The actual submission is done by a person in that system |
| Two-person approval | An action that requires approval by two people before it executes. The first approver cannot also give the second approval → Access Control |
| Automation (formerly Agent / Agent Builder) | A recurring check you delegate in a single sentence (/automations). On its own it only reads; writes are attached to the result card as drafts and approved by a person in Proposals. The name agent remains as-is in code and APIs → Automation |
| Send to (Automation) | Where an automation's results are delivered — app notification, company email, or messenger channel. It is fixed at creation time, and AI cannot change the recipient or address during a run |
| Notification channel | Teams and Slack channels registered by an administrator in the admin console. Selected by name in an automation's "Send to" → Copilot Admin Console — Notification Channels |
| Allowed mail domains | The list of domains automation email may be sent to. If it is empty, automation email is disabled |
| Dry run | A run performed before enabling an automation that actually calls the tools so you can confirm "this is what will go out." It creates no records, notifications, or cards. An automation that has not been dry-run cannot be enabled |
| Revision (Automation) | The version recorded each time an automation is saved. You can compare two revisions or roll back to an older one |
| Status dot | The only place color is used on screen — red for urgent, orange for caution, green for normal |
| Today / Home (former name) | The old start screen (/home). Now "Today's Tasks" on the Ask start screen |
| Anomaly Diagnosis (former name) | An old menu (/anomaly). Now "Anomalies to check" under Proposals › Tasks › Unplanned |
| Resolution (former name) | An old menu (/resolutions). Now the later stages of Proposals — there is no separate menu |
| Reports (former name) | An old menu (/report). Now Proposals › Completed |
| Watch (removed) | The old feature where you wrote down what to watch and received periodic checks. It has been removed — things to watch are now built as automations (the "Condition Monitoring" template). The old URL (/watch) does not redirect → Automation |
Where each task from the old screens now lives is documented in Copilot Screen Map — Old URLs.
ISA-95 / Asset Hierarchy
| Term | Description |
|---|---|
| Site | Plant/site — a physical location |
| Area | Area — manufacturing building, utility building, etc. |
| Line | Production line |
| Equipment | Equipment — fillers, packaging machines, compressors, etc. |
| Tag | Sensor tag — a measurement point such as temperature, pressure, or vibration |
| Point | Time series data point — an individual value collected from a sensor |
| OPC | OLE for Process Control — the industrial automation communication standard |
Production Metrics
| Term | Description |
|---|---|
| OEE | Overall Equipment Effectiveness — availability x performance x quality |
| Availability | Availability — actual run time against planned production time |
| Performance | Performance — actual output against theoretical output |
| Quality | Quality — the ratio of good units to total output |
| MTBF | Mean Time Between Failures |
| MTTR | Mean Time To Repair |
| MTTA | Mean Time To Acknowledge — time from alarm occurrence to acknowledgment |
| RAM | Reliability, Availability, Maintainability |
| EMS | Energy Management System |
AI / ML Terms
| Term | Description |
|---|---|
| RAG | Retrieval-Augmented Generation |
| LLM | Large Language Model |
| Embedding | The process of converting text or data into vectors (arrays of numbers) |
| Reranker | A model that re-evaluates the relevance of search results and reorders them |
| Ontology | Ontology — a knowledge structure that defines relationships between entities |
| Knowledge Graph | Knowledge graph — a structure representing entities and relationships as a graph |
| Anomaly Detection | Anomaly detection — detecting data that deviates from normal patterns |
| IBM Granite TSPulse | IBM's foundation model for time series anomaly detection |
| TinyTimeMixer (TTM) | IBM's lightweight time series forecasting model |
| SSE | Server-Sent Events — real-time data transmission from server to client |
| Fine-tuning | Fine-tuning — further training a pretrained model on domain-specific data |
Documentation
| Term | Description |
|---|---|
| SOP | Standard Operating Procedure |
| MSDS | Material Safety Data Sheet |
| Docling | IBM's document parsing library |
| Chunk | A unit produced by splitting a document into fixed-size segments |
Infrastructure
| Term | Description |
|---|---|
| LiteLLM | A proxy server that unifies multiple LLMs behind a single OpenAI-compatible API (the chat LLM gateway). Deployed as the in-house image plantpulse-litellm wrapping the upstream (v1.103.0) |
| LightRAG | The official open-source server used as the RAG engine (v1.5.7) — hybrid search plus knowledge graph. Deployed as the in-house image plantpulse-lightrag wrapping the upstream |
| Qdrant | High-performance vector database |
| Valkey | Redis-compatible open-source in-memory cache |
| RustFS | S3-compatible object storage |
| DGX Spark | NVIDIA's compact AI computing system |
| Inbound key | A single key (PI_INBOUND_API_KEY) that validates requests arriving at the backend services (RAG, time series insight, document understanding, vision, Digital Thread). If it is empty the service will not start — the only way to run without authentication is the development switch PI_INBOUND_AUTH_DISABLED=true → Environment Variable Configuration |
| Request id | A value the backend attaches to each request, returns in the X-Request-ID response header, and writes into that request's log lines. It links error screens to logs |
| compose profile | A group of services excluded from the default startup and enabled separately — registry (air-gapped network image registry) and media (transcription plantpulse-asr · plantpulse-asr-audio, enabled by the PI_MEDIA_ENABLED switch) |
| Video/voice switch | A single host env line, PI_MEDIA_ENABLED=true. It governs, in one place, starting the transcription serving, letting rag-api accept video and audio, opening the transcription intake in document understanding, and pulling the transcription model. The former name MEDIA_INGEST_ENABLED remains an alias only through 2026.10 → Video and Voice Collection |
| One-shot service | A container for which finishing its work and ending in "Exited (0)" is the normal outcome — plantpulse-asr-audio (copies the transcription serving's audio libraries into a volume). bin/status.sh counts this as normal |
| ASR | Automatic Speech Recognition — transcription that converts speech into timestamped text. Used by video/voice collection and voice-based Ask. Serving is vLLM Whisper (service name aura-asr, model Aura-ASR-Turbo) and is disabled by default |
| Hardware profile | A file (compose/profiles/<모양>.env) holding the per-hardware-shape (gb10 · g7e) GPU shares and the limit on their sum. Selected by PP_HW_PROFILE in the host env |
Removed from the stack
In the 2026-07 revision, Langflow (unnecessary once chat orchestration moved inside Copilot) and Jupyter were removed.
⚠ Arize Phoenix is back. As of 2026-07-27 it has been reintroduced for agent turn tracing (OTEL) and is in operation → Monitoring
Terms from the Other Products
In PlantPulse, four products share the same asset hierarchy, tags, and points. Terms not found here are in the other three books — Platform Glossary · Edge Glossary · Studio Glossary.