AI assistant, agent and costs

Softic GMAO integrates artificial intelligence in two ways: a conversational assistant available to every user, and an analysis agent that prepares proposals for the planner. Both are measured and capped by a cost control.

The AI assistant

Ask the assistant (page buttons, or the Softic AI button in the top bar) opens a conversation. The assistant has reading tools on the CMMS: assets and meters, work orders and requests, plans, stock, permits, shutdowns, handover projects, KPIs, pending approvals, and this manual (search_manual, get_manual_chapter) to answer "how do I" questions. It can also, if you explicitly ask, create a work request or record a reading; it confirms it to you.

Examples: "Which work orders are overdue on U-100?", "What is the site's availability over 12 months and which equipment weighs the most?", "How do I isolate a permit?", "Create an urgent request on P-201A: leak at the stuffing box".

The Ai.ShareFreeText setting controls whether free texts (descriptions, notes) are sent to the model; the data is only used to answer, it is not kept by the provider beyond the request (see the Anthropic terms). Conversations are stored in the CMMS for the user.

The AI agent

AI agent shows the agent's proposals, run every night (gmao-ai-agent) or on demand (Run now). The agent prepares facts (worst-performing equipment and their plans, recurring SCADA alerts, parts below reorder point, backlog), then reasons with the reading tools and files typed proposals:

• Adjust a preventive plan (proposed interval, justification: failures, MTBF, MTTR);
• Create a work request (equipment, priority, description);
• Reorder a part (quantity, notification to the planners);
• Information (observation without action).

Each proposal carries its justification, its estimated impact and its priority. Nothing is applied without you: Apply executes the action (plan change, request creation with its approval circuit, notification), Ignore files it; the author, the date and the result are kept. The Agent.MaxRounds, Agent.MaxTokens and Agent.MaxProposals guardrails bound each run, hence its cost.

AI costs

Administration > AI costs measures every call (assistant, agent, MCP): user, model, input, output and cache tokens, valued according to the price grid and converted into a billed amount (configured service margin and exchange rate). The page shows the current month and the history, per day, per feature, per user and per model, with an Excel export.

• Monthly budget (Ai.MonthlyBudgetUsd): alert to administrators at Ai.AlertPercent (80% by default), notification on overrun; if Ai.StopWhenExceeded is active, the assistant and the agent stop until the next month or until the budget is raised.
• The Softic role also sees the raw cost and the margin; the client's administrator sees the tokens and the billed amounts.
• The model, the maximum response length and extended thinking are set in Administration > Settings (AI assistant section).

MCP server (Softic.AI integration)

The API exposes the CMMS tools through the MCP protocol (POST /services/mcp), protected by a key generated in AI costs. An enterprise assistant (Softic.AI) can thus query the CMMS — and read this manual — without direct access to the database. Calls are counted and billable per unit (Ai.McpCallPriceUsd).