Applies to: Pulse Copilot · edge/on-prem or managed cloud
Status: DRAFT — product copy grounded in shipping agent capabilities. For human review before external use.
Last reviewed: 2026-07-12
Sources:website/pulse-website-mockup_V10.html(AI-02 / AI-03 use-case framing); Agents architecture (actual sub-agent capabilities).
Pulse Copilot turns a plain-English question into an answer backed by real plant data. You ask; it retrieves the relevant tags, analyzes them, and replies with the trends and context — powered by the Clarity AI engine, with no query language and no data specialist in the loop.
Every use case below maps to a capability that ships today. Where a question needs a specific agent, this page names it and links to the Agents architecture for the exact behavior.
The value is the collapse of a multi-step, multi-tool afternoon into a single question:
| Today — hours lost | With Copilot — minutes |
|---|---|
| Equipment trips → check DCS trends → download data from the historian → analyze → check maintenance records → call an expert → find an answer and act | Equipment trips → ask a question → get the answer |
| Seven steps, several tools, one specialist | Three steps, one chat |
| Ask | What Copilot does | Handled by |
|---|---|---|
| "Why did the ID fan trip on boiler-2?" | Validates the incident against the live historian, examines the deviating tags, generates diagnostic plots, and adds historical-recurrence context, then synthesizes an executive diagnosis. | incident_agent (analyzer → assessment) |
| "Which sensors look like they're misbehaving on this unit?" | Detects open incidents per unit/system and surfaces the deviating tags with short- and long-term plots. | system_health_agent |
| "What questions should I be asking about this event?" | Generates targeted diagnostic questions to guide the investigation. | question_agent |
| Ask | What Copilot does | Handled by |
|---|---|---|
| "Show me steam pressure and feedwater flow for boiler-2 over the last 24 hours." | Resolves the tags, fetches the series from the historian, and returns the values (and can chart them). | meta_data_agent |
| "How volatile has idfan_vibration been this week?" | Fetches the series, then runs statistical analysis in a sandbox to compute the summary. | meta_data_agent → data_analysis_agent |
| "Build me a dashboard for this unit's key tags." | Recommends dashboard sections across five visualization types — lines, histogram, scatter, table, comparison. |
dashboard_creation_agent |
| Ask | What Copilot does | Handled by |
|---|---|---|
| "How did the plant perform versus last month?" | Retrieves the period data and analyzes it for the comparison. | meta_data_agent → data_analysis_agent |
| "What are the major issues I should know about today?" | Lists current incidents for the unit (dropping noise) and can diagnose the significant ones. | meta_data_agent (getIncidents) → incident_agent |
All examples use the wiki's canonical dataset — org
acme-power/ siteplant-1/ unitboiler-2, tagssteam_pressure,steam_temp,feedwater_flow,idfan_vibration. See Data model & scope.
"Why is idfan_vibration rising on boiler-2?"
Everything the answer asserts comes from those tool calls — see Boundaries below.
Being explicit here is what makes Copilot trustworthy on a plant floor:
meta_data_agent has already retrieved — so answers trace back to real, fetched series.CLARITY_APP), the system-health and incident-summary tools are excluded; data retrieval, analysis, dashboards, and questions remain. Plan role expectations by deployment (details).You decide what leaves the plant network. Copilot is built to run where your data already lives — bundled with the historian at the edge, air-gap capable — so adopting it does not change your OT security posture. See the Security Model.
[VERIFY] The exact offline-AI architecture (which model runs where, which features work fully air-gapped) is pending a canonical definition — do not assert offline-AI specifics externally until confirmed. See also the local-inference roadmap scoping (
product/roadmap/local-llm-granite.md, local draft).