AI analysis for Ignition 8.3

Catches the drift before the stop. Answers the question before the meeting.

Artekis is an Ignition module. Every hour it screens your process measurements against their own normal for the machine’s current state, and turns the anomalies your engineers confirm into alarms. Any time, your supervisors ask about press performance, scrap or alarms in plain English and get an answer from the data already in Ignition, with a citation for every claim.

Contact us → Try the live demo →
What every finding carries
  • What and whereThe asset and tag or alarm path it is about, with a severity.
  • The evidenceTrends, counts and timelines from your own gateway data.
  • The proposed changeWritten in Ignition’s own property names, on the UDT type when the alarm comes from one.
  • Do nowThe action for the people on shift.
  • Your verdictConfirm, known issue, dismiss or wrong, recorded with a name.
Artekis in a Perspective session: the jet-cool finding on press DCM-302 with what was found, why it matters, what to do now and the evidence table, and the Ask panel answering the finding’s own follow-up question
Artekis on our demo plant: the jet-cool finding on press 302 with what was found, why it matters, what to do now and its evidence, and the Ask panel answering “what is the likely cause, what should I do about it, and how do I check the fix worked?” Click to open full size.
See it run

Four minutes on a running die casting plant

Recorded on Whitfield Die Casting, our demo plant: twelve die casting presses, two melters and three machining cells on Ignition 8.3, with the same Artekis module you would install.

Weeks of warning, not seconds1:10A press’s hydraulic oil climbs toward its alarm for two weeks and nothing fires. Anomaly Watch flagged it on the first day of the climb, in time to clean the cooler on a planned stop. Set up from one sentence.
Ask your plant1:14Scrap jumped on a press with no alarm. The supervisor asks the press, and the answer names the setpoint change, who made it and the data behind each claim, then looks ahead to the next press heading for trouble.
The likely cause, a minute after the alarm0:52A jet-cool alarm goes active on a press. Under a minute later, the likely cause and what to check first are on that press’s own screen.
What your alarms say about themselves0:56A safety guard bypassed for three weeks, six in ten alarms marked High, a cooling tower alarm that chatters. Alarm Health finds each one and writes the fix.

Try it yourself on a running plant

Whitfield Die Casting, our demo plant, runs Artekis on Ignition 8.3 around the clock. Browse its findings, open one to see the evidence, and ask about any press or machine in plain English. No sign-in, on a computer or a phone; signed-out visitors share a daily allowance of questions.

Open the live demo →
For the plant manager

One list of what needs attention, with the proof attached

Every analysis writes to the same findings page in Perspective. Reading it takes no Ignition knowledge and no trend work: each row is a problem in plain words, with a severity and the machine it is about. Open one and it shows what was found, why it matters, what to do now and the evidence behind it, drawn from your own gateway data.

Start the meeting from evidenceFilter by severity, area or status and the morning meeting starts from what the data shows changed, not from who remembers what.
Know who decided whatYour engineers confirm a finding, mark it a known issue, dismiss it or call it wrong, and each verdict is kept with a name and a note.
Resolved means the data agreesA finding resolves itself when the analysis stops seeing the problem, so a fix is shown to have held instead of being reported as done.
Reaches people who never open itHigh-severity findings can go out as an email digest, and the open-finding count and highest severity are tags you can show on any screen.
For example On our demo plant, a jet-cool alarm went active on press 302. The finding names a likely restricted line, most often scale in the strainer basket, because the flow fell from about 3.9 to 2.5 gpm while the die ran about 25 °F hotter in the same sample. Its Do now follows the plant’s own procedure: check the cooling tower’s supply pressure first, and if only this press is low, clean the strainer and restart at 3.6 gpm or more. A manager reading the list sees the problem, the cause and the fix in one row, and the evidence is a click away.
The Artekis findings page in Perspective: findings with severity, status, the finding in plain words, the asset and when it was last seen
The findings page: every analysis in one list, newest first, with filters for severity, status, playbook, asset, text and dates.
One Artekis finding: what was found, why it matters, what to do now, and the evidence trend
One finding: what was found, why it matters, what to do now, and every piece of evidence behind it.

Open the findings page yourself

The demo plant’s findings page is open to visitors. Pick any finding to see its evidence, then press Ask about this to ask a follow-up question.

Open the findings page →
Why it matters

The data is already there. The time to read it is not.

An Ignition gateway holds years of tag history, every alarm that ever fired, every setpoint change and every operator action, and usually the production database beside it. Nobody has time to watch a thousand trends for the one that started drifting on Tuesday, and the questions a supervisor asks at the morning meeting mean a controls engineer pulling trends and journal exports for an afternoon. Artekis does both: it watches the measurements every hour, and it answers the questions with evidence a person can check.

Anomaly Watch

A chat box answers what you ask. This finds what you did not know to ask.

Anomaly Watch is the part no chat assistant can do. Every hour it compares each historized measurement with its own baseline for the machine’s current state and product, so running torque is judged against running torque on the same part, never against a weekend average. When something drifts, sits outside its band, sticks, swings or flickers, it files the anomaly with the numbers. When an engineer confirms one, Artekis proposes the alarm that would have caught it, replays that alarm over months of your own history to score it, and hands over a tag import. Nothing is trained, and nothing is written to your gateway.

For exampleOn our demo plant, press 205’s hydraulic oil climbed from 118 °F to 132 °F over two weeks. The alarm is set at 140 °F, so nothing fired. Anomaly Watch flagged the drift on the first day of the climb, at about 121 °F, against that press’s own running baseline. That is roughly three weeks of warning to clean the oil cooler on a planned stop, instead of losing the press for a shift. For a plant this size we estimate a lost press-shift at about $15,000 of production.
The Anomaly Watch review queue: anomaly cards with the trend, the span shaded, the numbers, and Real issue, Not an issue and Known buttons
The review queue on our demo plant: two drifts in press 304’s intensifier response as its shot sleeve wore, each with its trend, the span shaded, the figures behind it, and one-click verdicts.
1 Screen
Every hourFive checks over each measurement, against a per-state baseline from the last 30 days.
2 File
With the numbersEach anomaly is stored as an event with the check’s own figures and a trend.
3 Confirm
Your engineerReal issue, known, or not an issue. A verdict takes one click in the review queue.
4 Score
Against historyA confirmed anomaly proposes a detector, replayed over 180 days: fires, real issues matched, false alerts.
5 Import
As a real alarmOne expression tag and its alarm, in Ignition’s own import format, for your engineer to import.

It also checks the historian itself: gaps, duplicate samples and sample rates that changed, reported to whoever owns the gateway. Anomaly Watch in detail →

Real questions

The questions your team already asks

Each one is answered from your own gateway. Every claim in an answer carries a citation to the data it came from, and an answer that cannot cite its source says so instead of guessing.

Why was press 4’s cycle time up on second shift last night?
The press’s history against its own normal for that product and state, and the audit log for any setpoint or recipe change in the window.
The answerCycle time averaged 41 s against a 37 s normal for this part. The audit log shows the clamp pressure setpoint lowered from 1,850 to 1,700 psi at 22:14 by the second-shift lead; cycle time rose from the next shot.
Ask
What changed on line 2 before the stop at 2:14?
The 30 minutes before the stop across the line: the first-out alarm, the first tag outside its normal range, an operator write, a configuration change or a lost connection, ranked.
The answerFirst out was the case packer’s infeed jam at 2:13:38, 22 seconds after the upstream conveyor’s photo-eye count went flat. The 2:14 line-stop alarm is a consequence, not the cause. Do now: clear the photo-eye on conveyor 2B.
Incident
Is scrap trending up, and on which machine?
Your production database, through a named query your team approves. Artekis never runs free-form SQL, so IT decides what the model can see.
The answerScrap rose from 2.1 % to 3.4 % week over week. Seventy percent of the increase is cell 3’s porosity rejects on part 1142, all on second shift, starting the day after the die was changed.
Ask, named query
Which alarms are wasting the operators’ time?
Chattering, fleeting, stale, duplicate and mis-prioritized alarms measured against ISA-18.2, each with the configuration change that fixes it.
The findingSump high-level activated 412 times last week, 3 to 5 times a minute whenever the pump cycles. Proposed: a 15-second clear delay and a 2 % deadband, on the UDT type, reaching all 14 sumps.
Alarm Health
Which measurements are drifting from normal, and since when?
Every historized measurement screened against its own baseline for the machine’s current state and product, hourly or daily.
The findingChiller 2 condenser approach temperature has climbed 1.1 °C a week for three weeks while running, now 3.4 standard deviations above its baseline. No alarm limit catches a slope. Confirm it to get the alarm that would have.
Anomaly Watch
What went wrong on the gateway overnight?
Log bursts, failing scripts, a Store and Forward backlog, dropped devices and bad-quality tags, compared with the day before.
The digestThe ERP sync script began failing at 01:12 after 3,000 clean runs: connection refused. Store and Forward holds 38,000 rows. The boiler PLC reconnected 46 times overnight; its status reads Connected this morning.
Gateway Health

Ask also searches the manuals and SOPs you upload, cites them by passage, and can draft the Anomaly Watch setup for an area from one sentence. Everything it answers from, it reads through a fixed set of read-only tools. The scenarios on this site are drawn from the plants we work in: foundries, machine shops, water utilities, data centers and packaging lines. The numbers are illustrative, not customer records.

What it does

Five analyses, one module

Ask runs when someone asks. The other four run on a schedule or when something happens, and land as findings your engineers review.

Hourly or daily

Anomaly Watch

Screens process measurements against their own history for the machine’s current state: drift, out of band, stuck, fast swings and flapping quality. A confirmed anomaly becomes an alarm you can import.

For exampleA flow transmitter reads exactly 412 gpm for three hours while the pump cycles on and off. The totalizer and the chemical dosing were wrong all morning. Stuck is filed within the hour.

Anomaly Watch in detail →

Weekly

Alarm Health

Finds nuisance, stale, duplicate, mis-prioritized and flooding alarms against ISA-18.2, proposes the configuration change that fixes each, and builds the alarm KPIs for a dashboard and a PDF report.

For exampleOperators at a water plant acknowledge 2,400 alarms a week, and 60 % of them come from eleven chattering level alarms. The proposed clear delays and deadbands take the week under 400 without touching a setpoint.

Alarm Health in detail →

Any time

Ask

A question about a plant area, answered from live tags, history, alarms, the audit log, your documents and your named queries, with a citation for each claim. Docked beside your Perspective screens.

For exampleA shift supervisor asks why scrap went up on press 203 this week, and gets the setpoint change, who made it and when, cited to the audit log and the tag history, without opening a single trend.
When an alarm fires

Incident

The likely cause of a high-priority alarm or a line stop, ranked from the minutes before it, with the action to take. The headline reaches the asset’s tags in about a minute.

For exampleA mixer trips on high current. The 30 minutes before it show the discharge valve’s open feedback dropping out 40 seconds earlier, right after a maintenance login on that PLC. Likely cause: the valve feedback, not the motor.
Daily

Gateway Health

What got worse on the gateway in the last 24 hours: bursts of errors in the logs, failing scripts, a Store and Forward backlog, connections and devices that dropped, and tags that went bad.

For exampleA tag-change script that bills downtime started throwing errors at 2 am after a database password rotation. Nobody noticed, because the screens still rendered yesterday’s numbers. The digest names the script and the first failing minute.
Yours

Your plant’s knowledge

A context model of each area, drafted from your tag tree in seconds: assets, units, normal ranges, alarms, your engineers’ rules, and the manuals and SOPs you upload. It is what tells a run that a pressure tag is the inlet of pump 3.

For exampleA rule written by your metallurgist, “if quench tank temperature exceeds 45 °C, parts from the last 10 minutes go to re-inspection”, becomes part of every Incident and Ask answer about that cell.
How a finding is made

Gathered on the gateway, explained by the model, decided by your team

The model never goes looking through your system. Code on the gateway decides what a finding can be about, and the model explains it.

1 Gather
On the gatewayAnalyzers read the alarm journal, tags, history, logs, audit records and your named queries.
2 Narrow
Fixed candidatesThe analyzers produce the list of things a finding is allowed to be about.
3 Explain
Your modelThrough your key, the model writes the findings from that evidence.
4 Check
Rule by ruleEach answer is validated, and anything that fails a rule is dropped and logged.
5 Decide
Your engineersFindings are stored with their evidence in your database for your team to review.

Findings are deduplicated across runs, resolve themselves when the analyzers stop producing them, and keep your engineers’ verdicts. How it works →

Inside Ignition

Results where your team already looks

  • Perspective views. Findings, finding detail with every piece of evidence, alarm KPIs, the anomaly review queue, an incident card, and an Ask panel docked beside your screens. They use your theme.
  • A PDF Alarm Health report through the Reporting module, ready to schedule.
  • An email digest through a gateway email profile.
  • Tags and MQTT for open findings, severities and incident causes, and a scripting API, system.artekis.
In your control

It recommends. It never acts.

  • Read only. It never writes to a device, setpoint or alarm, and never applies a change.
  • Your key, your model. Stored in Ignition’s Secret Provider and never shown, and a local model keeps everything on your network.
  • Your database. Runs, findings and verdicts stay in your SQL Server, MariaDB, MySQL or PostgreSQL.
  • Every call recorded with its cost, and a free preview of exactly what the model would see.

Security and control →

What it costs to run

Model calls measured in cents

You pay your model provider directly, and Artekis shows every dollar on its Cost page, with daily caps per provider, per run and per user. On our acceptance run, against a 125,000-event alarm journal and a 19,900-tag provider:

About a minutefor a 30-day Alarm Health run, under $0.50 in model calls
About 20 secondsfor a 24-hour Gateway Health run, about two cents
$15 to $24 a monthfor a gateway producing 30 to 50 alarm candidates a day; a small gateway a few dollars
$10,000$7,000one-time, per gateway, engineer included2026 price, through December 31, 2026

One license covers every analysis on one gateway, with annual support for updates, the same model as Ignition. It includes four hours of an Artek engineer’s time to install it with your team, draft your plant model, run your first reports, and show your supervisors how to ask questions it answers well. There is no cloud service to subscribe to; model usage is paid to your provider, in cents.

Pricing in detail →

Ways to start

Ask, set it up, or try it

Try it yourself on a running plant

Whitfield Die Casting, our demo plant, runs Artekis on Ignition 8.3 around the clock. Browse its findings, open one to see the evidence, and ask about any press or machine in plain English. No sign-in, on a computer or a phone; signed-out visitors share a daily allowance of questions.

Open the live demo →
One message

Ask us

Tell us about your gateway and the questions your supervisors ask about the plant. A member of our team reads every message and replies.

Contact us →

Included in the license

Guided setup

An Artek engineer installs Artekis with your engineer, drafts the plant model and runs your first reports together, within the four hours the license includes.

About guided setup →

Two-hour trial

Run it on your gateway

Artekis runs in Ignition’s two-hour trial, resettable like any module, and is free on Maker Edition for non-commercial use.

Ask for the module →

Requirements

What it runs on

Ignition 8.3.4 or later8.3.8 is the verified platform, with Perspective. Not available for 8.1.
A databaseSQL Server, MariaDB, MySQL or PostgreSQL, through a gateway connection. Anomaly Watch also needs tag history on the measurements it screens.
A modelAn Anthropic or OpenAI key, or an OpenAI-compatible endpoint, hosted or on your own network.
Built by Artek

From the team that builds Ignition systems

Artek is an Ignition integrator and an Inductive Automation Premier Integrator, founded in 2020 by former Inductive Automation engineers. We build the kind of Ignition systems Artekis analyzes, for manufacturers, water utilities, data centers and energy companies.

Visit artekis.io →
Questions

Common questions

Does Artekis change anything on our gateway?+

No. It never writes to a device, OPC item, setpoint or alarm, never acknowledges or shelves an alarm, and never applies a configuration change. A proposed change is a recommendation your team applies under its own change control.

Can it answer questions from our production database, like scrap or downtime?+

Yes, through named queries. Your team writes a select-type named query in a project Artekis is allowed to read, and Ask can call it, up to 1,000 rows at a time. Artekis never runs free-form SQL, so your IT team decides exactly what the model can see.

Which AI model does it use?+

One you choose, through your own key: Anthropic, OpenAI, or any OpenAI-compatible endpoint, including a model you run on your own network. Every call is priced and recorded.

We were asked about Microsoft Copilot. How does Artekis compare?+

Copilot works on Microsoft 365 data and cannot see an Ignition gateway. Artekis runs inside the gateway and reads your tags, history, alarms, audit log and production database directly, then sends only the evidence for one question or one run to the model you choose. It answers in Perspective, where your supervisors already are, with a citation for each claim.

Which Ignition versions does it support?+

Ignition 8.3. The module loads on 8.3.4 and later, and 8.3.8 is the verified platform. Ignition 8.1 is not supported.

How is it licensed, and what does it cost?+

A one-time license per gateway, normally $10,000 and $7,000 for purchases through December 31, 2026, activated through Ignition’s own licensing, with an annual support plan for updates and new releases, the same model as Ignition. The price includes four hours of an Artek engineer’s time to set it up with your team. You can evaluate it in Ignition’s two-hour trial first, and it is free on Maker Edition under Maker’s non-commercial terms.

Next step

See it on your own data. Get in touch.

Tell us about your gateway and what you want to find out. A member of our team reads every message and replies.

Contact us →
info@artekis.ai