Skip to content

AI Technician

The AI Technician component, in the AI Agents group, has Profinity AI review the profile and publishes the verdict as tags. It needs the AI licensed feature, included in the Server and Enterprise editions, and Profinity AI configured under Profinity AI settings. Each review is a round, and a round starts on a wake: a schedule, the Run AI Technician action, or an alert. For what happens during a round, refer to How an AI Technician works.

flowchart LR
    Profile["Profile"] --> Technician["AI Technician"]
    Technician --> Findings["Findings tags"]
    Findings --> Rules["Findings rules"]
    Findings --> Historian["Historian"]
The technician reads the profile and publishes findings tags, which the findings rules and a historian use like any other tags

Quick start

You need:

  • Profinity AI set up and turned on, as described in Profinity AI settings.
  • An enabled user for the technician to run as, usually a service account.

  • The Allow component actions permission, for step 5.

  • Select ADD COMPONENT, then select AI Technician in the AI Agents group.

  • In Run as, choose the user.
  • On the Instructions tab, replace the example instructions with your own review, as described in Write instructions.
  • Select ADD COMPONENT.
  • Select Run AI Technician from the technician's menu.
  • When the round ends, check that the Outcome findings tag reads attention or clear.

The full procedure, with a schedule, is in Set up an AI Technician.

Settings

Every setting is on the Settings tab, except Instructions, which has its own tab.

Setting Type Default Effect
Name Text AI Technician The component name. Rules name the technician by this name when they use it as an action.
Enabled On or off On When off, every wake is dropped and no rounds run.
Run as Dropdown of enabled users, required None The user each round signs in as. Its roles limit what the technician can see.
Instructions Text, required, full width An example industrial drive review Instructions for the technician. Refer to Write instructions.
Log output On or off Off Also writes the round log to the Profinity log. Dropped wakes and blocked rounds are written to the Profinity log whether this is on or off.
Schedule No schedule, Run on time interval, or Run on CRON schedule No schedule When scheduled rounds run. With No schedule, rounds start only from Run AI Technician and from alerts.
Time interval and Time interval unit Whole number and a unit Empty, seconds Shown when Schedule is Run on time interval.
Cron schedule Text Empty Shown when Schedule is Run on CRON schedule.
Round timeout (seconds) Whole number, 30 to 3600 300 A round running longer than this stops with a blocked finding.
Alert cooldown (minutes) Whole number, 0 to 1440 60 Alert wakes are dropped for this long after a round ends. 0 turns the cooldown off.
Daily limit Whole number, 0 to 1000 10 The most alert and Run AI Technician rounds in one day. Scheduled rounds do not count.

A Quartz cron expression has six or seven fields that start with seconds, so 0 0 0 * * ? runs at midnight every day. Run scripts describes the same format. Profinity rejects a save with an expression it cannot read, and it reads the expression on the clock of the computer that runs Profinity.

Example

This example runs a review every night at midnight as a service account named technician. Add it under components: in the profile.yaml file of the profile.

- component:
    type: AiTechnicianComponent
    id: AiTechnicianComponent1
    name: AI Technician
    tagTreeParentPath: /
    enabled: true
    runAsServiceAccount: technician
    instructions: YOUR_INSTRUCTIONS
    scheduleMode: CronSchedule
    cronSchedule: 0 0 0 * * ?
    timeoutSeconds: 300
    alertCooldownMinutes: 60
    dailyLimit: 10

Replace technician with the user name of the account the technician runs as, and YOUR_INSTRUCTIONS with the review. The keys match the settings: runAsServiceAccount is Run as, scheduleMode is Schedule, timeoutSeconds is Round timeout (seconds), and the others follow the setting names.

Actions

The technician has two actions, which need the Allow component actions permission:

  • Run AI Technician starts a round now. It is available while no round is running, and it counts towards the daily limit but not the cooldown.
  • Cancel Run stops the round in progress, which ends with a blocked finding and the message Round was cancelled.

Findings tags

The technician publishes these tags under its Findings branch. Rules, dashboards, and historians use them like any other tag.

Tag Values
Outcome attention, clear, or blocked
Severity 0 for none, or 1 (Trace), 2 (Debug), 3 (Info), 4 (Warning), 5 (Error), 6 (Fatal)
SeverityName The severity as text, such as Warning, or empty for none
Message A short summary of the finding, or the reason a round was blocked
Wake schedule, manual, or alert, the wake that started the round
RoundTime The time the round finished, in UTC
Report The round report: the wake, and the AI's visible reply
FailedRounds The number of blocked rounds in a row, reset to 0 by an attention or clear finding

The technician also publishes RoundStatusMessage, which reads running, attention, idle, or blocked, and NextWakeMessage, which holds the next scheduled round in UTC or none.

Profinity saves the last finding and restores it when Profinity or the profile restarts, so an alert raised from a finding stays raised after a restart. A new technician has no finding, so its findings tags have no value and no findings rule fires until the first round ends.

Built-in rules

The technician ships three rules, which you open and change with Edit Rules on the technician, as described in Tag rules. None of them ships an action. Add the actions you want, such as an email at Warning and above.

Rule Raises
Findings severity An alert at the level of the Severity tag, from Trace to Fatal. It clears when a clear finding sets Severity to 0.
Findings message A Warning when the Message tag changes and the outcome is attention. The same text does not send again. A clear finding ends the alert. A blocked round leaves it raised and does not send.
Technician not working A Warning when FailedRounds reaches 3, at most once an hour.

Findings message sends each distinct attention finding, even when two rounds give the same severity. Its On update setting is on, so it fires again while its expression is already true, as described under On update. The shipped expression sends only attention rounds. To send blocked or clear rounds as well, edit that expression. Its related tags name the other findings tags, so the expression and the action message can read them, as described under Related tags.

Refer to Rule actions for the actions a rule can run.

Dashboard

The technician dashboard needs the View tags permission. Its status pill shows STATUS, from RoundStatusMessage, and NEXT, from NextWakeMessage. The Findings panel shows Outcome, Severity, Message, Woken by, and Round time.

The Last round panel shows the round log while a round is running: the round start, each step, the finding, and the round end. The log is cleared when the next round starts, and a dropped wake is added to it. The log is held in memory, so it is empty after Profinity restarts. When no round is running, the panel shows the Report tag instead, with its line breaks kept. A historian that stores text can show that report at a past time, as described in View a dashboard at a past time.

Keep a history of findings

To look back at past findings, record the findings tags with a historian, as described in How to configure data logging. Outcome, SeverityName, Message, Wake, and Report hold text. A TAG SQL Historian or TAG Local Historian stores text, and an InfluxDB historian stores only tags that hold a number.

Report is the largest tag, so include it in the logged collection only when you want the report kept for every round.

Troubleshooting

What you see Cause What to do
Outcome is blocked and Message reads AI is not licensed or configured or AI settings are incomplete Profinity AI is not licensed, or its settings are not complete Complete Profinity AI settings, and check the edition in Licensing.
Message reads Run as user was not found or is disabled Run as names no user, or a disabled user Set Run as to an enabled user. A service account is the usual choice.
Message reads Model did not call record_finding The AI replied without giving a verdict Make the instructions end with a clear instruction to give a verdict, as described in Write instructions.
Message reads Round timed out The round took longer than Round timeout (seconds) Increase the timeout, or narrow the instructions.
Run AI Technician or an alert does nothing Profinity dropped the wake. The Profinity log has a line AI Technician '<name>': dropped with the reason, which the table in Dropped wakes explains. Fix the cause for the reason in the log.
A finding names data the technician should have seen as missing The Run as account lacks the permission to view it Add the view permissions to the account's role, as described in Service accounts.
Text findings are empty when a dashboard is viewed at a past time The designated historian is an InfluxDB historian, which stores numbers only Use a TAG SQL Historian or TAG Local Historian as the designated historian.
Technician not working alert Three rounds in a row were blocked. Read the Message tag and find its text in this table.

Dropped wakes

Reason in the log Cause What to do
DroppedDisabled Enabled is off. Turn on Enabled.
DroppedBusy A round was already running. Wait for the round to end, then select Run AI Technician again.
DroppedCooldown An alert arrived within Alert cooldown (minutes) of the last round. Wait for the cooldown, or lower Alert cooldown (minutes).
DroppedDailyLimit The day's alert and Run AI Technician rounds reached Daily limit. Wait for the next day on the computer's clock, or raise Daily limit.
DroppedLoopGuard The alert came from a rule on the technician's own findings tags, which would start rounds in a loop. Leave the technician out of the actions of its own findings rules.

To have device alerts start a round, continue to Investigate alerts with an AI Technician.