Meet Fia — Fentrica Intelligent Agent

Fentrica's AI has until now been a handful of unrelated buttons — a suggestion here, an analysis there, each authored by nobody in particular. From this release they are one agent with a name, a face and an audit trail: Fia, the Fentrica Intelligent Agent.

Nothing has been taken away, and nothing new has started running on your sites without a switch. What changed is who does it: every AI-authored change is now attributed to Fia, who shows up in your activity feeds exactly where a colleague would.

Fia is a colleague, not a background process

Fia is a real member of your organization rather than an anonymous automation event. She appears in avatars, activity feeds and author lines through the same machinery as everyone else, and you can open her profile from Managers at both organization and site level to read exactly what she does.

• Named on every change. Each field she fills in and each analysis she writes leaves a change-log entry authored by Fia, so what she did and when is always on the record.
• She never overrules a person. A priority a manager has already set is never overwritten, and a category somebody chose by hand is never removed.
• She cannot be edited, removed or signed into. Fia is part of the platform; her profile is read-only and no login can ever be her.

Reading maintenance requests

When the request classifier is enabled, Fia reads each new maintenance request as it arrives and fills in what it is about, before anyone opens it.

• Categories from your site's own ticket tags — at most three, and only those she is genuinely confident about. Tags are added, never removed, so re-reading a request after an edit does not churn your data.
• Priority, judged by the consequence of delay rather than by how forcefully the request is written — and only when nobody has set one.
• Scope: one home, a shared space, or the whole site.
• The space affected, matched from the reporter's own words against your site's records. If the words identify more than one space, or none, she leaves it unset rather than guessing at a door.
• Safety hazards, in a deliberately narrow sense: a gas smell, fire or smoke, a carbon-monoxide alarm, exposed or arcing electrics, water actively flooding, or a person trapped. Nothing else is flagged as a hazard however urgently it is written — no heating, no hot water, a leak into a bucket and a stuck door are high priority at most.
• What the report is still missing, so somebody can ask for it instead of dispatching a technician who cannot finish.
• The language it arrived in. Fia reads each request in that language and never translates it in order to file it.

This is an assessment, not a dispatch: Fia does not assign the request to anyone, and she does not touch deadlines or service levels. A request a manager has already triaged is left alone entirely.

Answering the phone

On sites with a voice line configured, Fia takes the call herself — she asks what is wrong and opens the maintenance request without anyone picking up.

• The recording is transcribed onto the request, so the words are there to read rather than a recording to sit through.
• The transcript is the request's description, in the language the voice line is set up for.
• Fia waits for the transcript before reading the request. A call that has not been written up yet is not classified from its title alone.

Drafting replies to tenants

Fia's message helper both polishes a draft you have written and writes one from scratch, drawing on how similar requests at your sites were answered before.

• In the tenant's language. The reply goes out in the language that tenant wrote in, ahead of the site's default, and you can override it per message.
• Informed by the assessment. What she determined about the request shapes the reply — an urgent hazard is answered as urgent, a site-wide fault acknowledges that others are affected too.
• Polishing does not rewrite your intent. Asked to polish, Fia will clarify a question your draft already asks but will never add one of her own. She also never comments on how the reporter came across.

Analysing alarms

Ask Fia about an alarm and she gives you severity, the likely cause and what to check first. She reads more than the alarm line itself: the system's installed spare parts and their lifecycle state, that alarm's own history, related systems and — where you have enabled it — your property documents. Each of those context sources is a per-site setting you control, so you decide how much she is given to work with.

The analysis is saved on the alarm and logged in its own Log tab as a change by Fia, so an analysis never simply appears authored by nobody.

Forecasting meters and planning the battery

Point Fia at a meter you already have and she projects it forward — and where you have solar, a battery and metering, she plans the hours ahead against day-ahead prices. Two things are worth being precise about, because the obvious phrasing overstates them:

• Retrained every hour, not learning over time. Each hourly run fits a fresh model on up to three months of that meter's own recent readings and forecasts up to 48 hours ahead. It is a model rebuilt from your data every hour, not one that accumulates knowledge. Time of day, day of week, holidays and a weather feed can be added as inputs if you want them.
• The battery, and your controllers do the driving. Solar, battery and grid limits are solved together towards the goal you pick — lowest cost, most self-consumption, or best return — and the result is published as charge, discharge and idle periods that your connected controllers pick up and apply, inside the state-of-charge and grid limits you set. The battery is the only equipment this drives; other plant has no dispatch path.

Separately, for Swedish demand tariffs, Fia keeps your billed peak down by working out an hourly import limit from your tariff rules and the peak hours this month's bill is actually based on. That one is a calculation from measured consumption, not a prediction.

Turning Fia on and off

Fia is controlled per site, with three independent switches:

• Alarm analysis — on by default.
• Message helper — on by default.
• Maintenance-request classification — off by default, because unlike the other two it is billed per request classified. Turn it on for a site when you want it, and turning it back off stops the spend immediately, including on requests already queued.

A quieter interface, and text you can actually read

Fentrica has grown a great deal in the last year, and the interface grew with it — a little at a time, in whichever direction each feature needed. The same release takes the whole platform through one pass and settles it on a single visual system.

Less chrome, more contrast where you are reading, and one accent colour spent only where it means something. Nothing has moved, nothing has been renamed and no workflow has changed. It should feel like the same product with the noise turned down.

Text you can read

This is the change you are most likely to notice.

Secondary text across the platform — the description under a list item, timestamps, table headers, the small print inside a card — was set in a grey chosen to be as light as it could get away with. On a laptop in a bright office, "as light as it can get away with" turns out to be too light.

Everything in that rank is now noticeably darker, and small text at eleven and twelve pixels gained the most. Nothing changed size or moved position; it simply stopped being faint.

Organization themes

If you run Fentrica in your own colours, and particularly on a dark background, parts of the interface were unreadable: headings and list titles rendered in dark ink on a dark surface, while the lines beneath them stayed legible. The strongest text on the page was the text that disappeared.

The whole neutral palette now follows your own colours — headings, secondary text, card borders and dividers are all derived from the ink and background you configured, rather than fixed to the default light theme.

• Card borders on a dark theme are visible without being stark: a hairline that reads as a hairline, not a hard outline.
• Scrollbars, dropdown internals, date pickers and browser autofill follow your theme too, instead of staying stubbornly light against dark content.
• If you use the default colours, nothing has changed. The original values are reproduced exactly, so an organization that has not customised its theme sees the interface it saw yesterday.

Charts

Series could share a colour. Past six lines the palette ran out and began repeating, so two different meters or two different sites were drawn identically. Charts with many series now produce as many distinct colours as they need, and those with a small fixed set use a palette checked for colour-blind readers.

Charts were in the wrong typeface. Nearly every chart drew its axis labels in a default font rather than the one the rest of the platform uses. They now match their surroundings, with axis marks and legends in the same quiet grey as everything else and figures aligned so digits line up in columns.

Charts inside reports also carry a distinct line pattern per series, so a report printed or photocopied in black and white can still be read.

Printing

Printing any page other than a report used to produce a single sheet — usually the navigation sidebar and nothing else. That was true of every list, dashboard and detail page in the platform.

Printing now produces the full document from any page, with navigation and on-screen controls left out and coloured status marks retained, so a printed page still carries its meaning.

Maintenance requests move themselves along

A request still marked New becomes Open as soon as somebody is assigned to it and the request is edited or commented on. Requests already in progress, on hold, resolved or closed are never touched: this only ever moves a request out of New, never back into it.

Routine maintenance over a full year

A maintenance rule that repeats daily used to put every one of its occurrences into a single month cell, stretching the row until the year view was unusable.

Each month now shows the first few and collects the rest behind a count. Opening it lists the individual dates, so you can see which days are covered and jump to any one of them. Occurrences are listed in date order, which they previously were not.

Smaller things, now put right

• A clearer first screen. Managers signing in before choosing a dashboard now get a start screen that says what it is actually doing: the dashboard you pick becomes your landing page, and you can switch from the sidebar whenever you like. It previously invited you to choose "the dashboard you want to open today", which rather understated the commitment.
• Icons that were not there. Twelve icons across the platform rendered as nothing at all — a blank space where a symbol should be. They now appear.
• Analytics charts on a technical system have more room, and the date range above them scrolls with the page instead of staying pinned over the content.
• Knowledge base categories show which one is selected; both states were previously styled identically, so the selection was invisible.
• Dialogs no longer open with a heavy highlight around the close button.
• A technical system's header is readable on a dark theme; the site and unit line beneath the title used to vanish.
• Tables, badges, chips and tags share one set of shapes, borders and colours rather than several near-identical variants.