Product Release

AlarmIQ Just Got Faster (Again). Here’s What’s New.

Published August 4, 2026 · 6 min read

We shipped a batch of AlarmIQ updates this month. The headline: AlarmIQ now lives inside every PulseMQ Machine Dashboard. One login, one screen, one live model of the machine — no switching apps to ask what is going on.

Every Dashboard Is Now an AlarmIQ Screen

Before this release, AlarmIQ was a separate screen. You watched the dashboard when things were fine, and you switched to AlarmIQ when something felt off. That made sense at first, but it made asking a question about the machine a two-step motion — and by the time you got there, the moment was often gone.

Now AlarmIQ is embedded directly into every machine dashboard. The dashboard still shows what it always did (OEE, availability, current job, live counters), but there is now a persistent chat sidebar wired to the machine you are looking at. Ask “why is Line 5 down?” and the answer is grounded in the exact machine on the exact screen — not a generic model, not last week’s snapshot.

The single sign-on story got simpler too. If you can see the dashboard, you can talk to AlarmIQ. No second license, no second URL.

Faster Answers (Again)

We already benchmarked the previous release at sub-two-second median responses. This release cuts that further for a specific class of question we saw dominate real usage: “what happened.”

• Instant timeline lookups. Questions that resolve to a fixed time window (“what happened in the last 30 minutes”) now short-circuit around the LLM roundtrip when the answer is already in the local replay buffer. The result reads like an AI-authored sentence but is served in about the time it takes to hit a REST endpoint.
• Warm PLC context. AlarmIQ keeps a rolling snapshot of the last hour of PLC values per machine, so common follow-ups (“which interlock,” “which sensor”) do not re-fetch the same data. The average follow-up is now noticeably faster than the initial question.
• Slimmer prompts. Half the previous prompt size — more of the token budget goes to actual machine context, less to boilerplate. Answers are shorter and get to the point faster.

PLC-Level Explanations, Not Just Alarm-Level

The original AlarmIQ answered “which alarm is active and what does it usually mean.” That is useful, but it stops one step short of the answer most operators actually want, which is: why is the alarm active right now?

The new PLC-level explainer walks the tag chain. When you ask about an active alarm, AlarmIQ inspects the exact PLC conditions that gate it — interlocks, permissives, safety chains, upstream sensors — and reports the specific one that is not satisfied. A typical answer now reads like:

“Line 5 is waiting on one interlock. Circuit Breaker 4 is open. The rest of the safety chain looks good. Reset CB4 and the line will attempt an auto-restart within 3 seconds.”

That is a real answer — the kind you would previously have gotten from an experienced technician after they walked the panel. It is also grounded in the live PLC state, not a document.

Shared Alarm History Across the Plant

AlarmIQ used to know one machine at a time. Now it knows all of them together. Ask “did we see this on Line 2 last week” and it can actually check. Ask “which line has the most CB4 trips this month” and you get a ranked list.

This turned out to be the feature maintenance teams asked for hardest. Every plant has that one recurring nuisance alarm that costs an hour a week. Finding the pattern used to mean pulling three months of PLC exports and pivoting them in Excel. Now you ask.

Ask-the-Machine on Mobile

Mobile AlarmIQ was previously a stripped-down version — text-in, text-out, no dashboard context. This release brings full feature parity to the mobile app. If you can ask it from your desk, you can ask it from the plant floor with your phone in hand.

Practically, this means the person who is at the machine is now the person doing the diagnosis. That was already the workflow that worked. The tooling finally caught up.

Evidence Contract Enforcement Turned Up

AlarmIQ has always answered under the Evidence Contract: every claim traces back to a specific PLC read or a specific dashboard value. This release turns the enforcement up.

The new refusal pass rejects answers that do not name their source, even when the underlying claim would probably be right. That produces slightly more “I do not know because I could not read tag X” responses — which is exactly the point. A tool that admits when it cannot see is a tool you can trust when it does. Read the Evidence Contract for the three rules AlarmIQ answers under.

Under the Hood

A few things changed that you will not see in the UI but will feel:

• The replay buffer per machine grew from 30 minutes to 60 minutes. Ask about anything within the last hour and it is instant.
• The PLC snapshot refresh dropped from 5 seconds to 1 second on active machines — the “is it still down” question now reflects the current state, not the state from a few seconds ago.
• Model responses now stream. You see the first sentence in about 300ms instead of waiting for the whole answer.
• Every AlarmIQ conversation is logged with the exact PLC reads that grounded each answer. If you ever want to re-verify what the tool told you, the paper trail is there.

What’s Next

A few things we are working on that did not make this release:

• Multi-line diagnostic mode. Ask a question that spans multiple machines (“why is the whole line down”) and get a coordinated answer instead of a per-machine reply.
• Historical replay. Point AlarmIQ at a past event (a downtime yesterday, an alarm chain from last week) and ask questions as if it were live. The replay stack is ready; the query surface is being polished.
• Predictive prompts. AlarmIQ noticing that a pattern of PLC values usually precedes a known failure and offering to explain it before the alarm fires. This one is being validated carefully — a prediction that is wrong more than a small percentage of the time is worse than no prediction.