Trackary AI: what artificial intelligence does in your CMMS (and what it doesn’t)

Reading a machine’s nameplate

You take a photo of a machine’s manufacturer nameplate. Trackary AI reads the make, model and year from it, and proposes them to fill in the machine’s record.

An unreadable serial number is not made up: the field stays empty and you enter it yourself. Check each proposed field before saving the record.

Reading a VGP inspection report and your documents

The VGP inspection report (procès-verbal, PV) is the report the inspection body gives you after the inspection. You import it into Trackary as a PDF or a photo. You can also forward it by e-mail to your organisation’s drop-off address.

Trackary AI picks out the findings word for word, the hour meter reading, the dates, the report number and the name of the inspection body. Each line shows the excerpt of the document it comes from (with the page number for a PDF). So you can check it against the original.

From these lines, it proposes defects and the date of the next VGP. Nothing is written until you have approved it: the proposals wait in a “To approve” list. You accept them line by line, or all at once after checking them, and you can adjust a value before approving. An approved proposal is saved under your name.

Findings (réserves) are the defects noted by the inspector. To know what to look for in the report, see how to read a VGP report; for what comes next, how to clear the findings.

1. Import the report: PDF, photo or e-mail sent to the drop-off address.
2. Check each proposal with its citation (the excerpt of the document it comes from).
3. Approve the line, or adjust the value and then approve.
4. If something is wrong, choose “Undo import”: Trackary explains what will be undone, then deletes the defects and the VGP due date created by that import.
5. Then the app’s home screen may suggest “N findings to clear”, with a button to go to them.

Asking a question about your fleet

You can type a simple question to Trackary AI, for example: “Which machines are out of service?”. It answers from your fleet in Trackary, with a link to the record of each machine it mentions.

The chat answers and takes you to the records. It neither creates nor carries out anything by itself: you are the one who acts in Trackary. Its answers may be incomplete: open the record to check.

Manuals, voice, QR code and authorisation cards

Other document reads follow the same principle: Trackary AI proposes, you check and you approve.

• Manufacturer manuals: Trackary AI reads a machine’s manual and proposes maintenance plans, to be approved before they apply. To build these plans, see the preventive maintenance plan.
• Authorisation cards: it reads an employee’s operator authorisation card and proposes what it reads.
• Voice: you dictate instead of typing, and it transcribes your dictation into text.
• QR code: on the page opened by scanning a machine’s QR code, it proposes a description of the photo taken.

What it is useful for in practice

Three common situations in a fleet:

• After a VGP: instead of copying out the findings from the report, you check them with their citation and approve them.
• When a machine arrives: a photo of the nameplate instead of typing the make, model and year. If the serial number is unreadable, it still has to be entered.
• Afterwards: the app’s home screen may suggest “N findings to clear”.

What Trackary AI does not do

• No breakdown diagnosis: on a fault code you give it, it explains the code and proposes a finding or a job, for you to approve.
• It never acts without approval: nothing is written to your fleet until a user approves it.
• It does not carry out the VGP, does not replace the inspection body, does not clear findings and certifies nothing.
• It does not take a machine out of service by itself: stopping a machine remains the employer’s decision, even if the report states that use is prohibited.
• It can be wrong: a document read or an answer may be incomplete or inaccurate. Check before any decision, especially on safety and compliance.

Data and privacy

Trackary AI uses language models from OpenAI, a company based in the United States. OpenAI is listed among the sub-processors in Trackary’s data processing agreement.

Neither The Level Code, the publisher of Trackary, nor OpenAI uses your data to train models. Apart from voice transcription, Trackary asks OpenAI not to store the responses. OpenAI may keep the exchanges for up to 30 days to detect abuse, then deletes them. A conversation with Trackary AI is deleted 12 months after its last message.

The administrator can switch Trackary AI off for the whole organisation or for certain roles, in Settings › Trackary AI; an administrator is never switched off by a role setting, only by the general switch. When Trackary AI is switched off for the organisation, no AI function runs: documents dropped off in batches or received by e-mail stay on hold and are analysed once AI is switched back on, photos taken from a machine’s QR code are saved without a suggested description, and the Excel import runs without AI suggestions. A role that is switched off cannot start any AI reading. The details are in article 11 of the terms and conditions.

What is included

Trackary AI is included in the subscription, with no paid add-on. Trackary costs €14 excl. VAT per active machine per month (€11 above 50 machines), unlimited users, with a 30-day free trial, no credit card required.

If a document read is refused, the document stays dropped off and you can restart its read later.