Guided and AI-assisted entry

The forms for a new asset, ticket or document have a side panel with two parts.

The checklist

Always there, no AI needed. As you type, it shows:

• the steps done so far, and the Next question worth answering;
• what is wrong or worth knowing, with a one-click fix where there is one: a key already taken (with a link to its record), a serial or inventory number another record holds, a likely duplicate, a control-channel name typed as if it were equipment, a missing required attribute;
• for a ticket: when an incident happened, which equipment it affects, similar open tickets;
• for a document: the code it will get, similar titles, the records it mentions.

Describe it (AI)

Appears when the workspace’s AI endpoint is set up (AI endpoint).

1. Write what you know, in your own words and any language (“Agilent ion pump VacIon Plus 75, serial 77120, in the gun area”), or use Add a photo or file: a nameplate photo, a PDF datasheet, a Word or Excel file, an email.
2. ARGUS shows suggestions, each with how sure it is and the words or page it was read from.
3. Press Use on the ones you want, or use all of them in empty fields. Nothing you typed is overwritten, and nothing is saved until you save the form.

For a ticket, possible causes are shown as unresolved hypotheses, never written as findings.

For a document, the panel starts from the document itself: paste its text or add its file, then Rework with the template or Keep it as it is (see Documents). The file is attached to the document, and the summary the assistant writes goes in the document’s Summary, never in place of its body.

What is filtered before the AI sees it

Before text is sent to the model, ARGUS replaces what looks like a secret with [redacted]: private keys, password=…, token=…, api_key=… and similar assignments, Bearer tokens, AWS access keys, GitHub tokens and user names with passwords inside web addresses. A value that looks like a password is never proposed for a field. The filter recognises these patterns, not every secret: still remove credentials and unrelated personal information before you paste text or attach a file.

Check suggestions before using them

Read the words or the page each suggestion was read from. Check identifiers, serial numbers, units and types. How sure it is says how confident the model is; it does not make a value right.

A datasheet describes a model’s range; it does not give the serial number or the settings of your unit. Use fills the form; Save creates the record. For an existing asset completed from a file, accepting the proposal in the Review queue is what changes the record.

Complete an existing asset from a file

On an asset’s page, Complete from a file takes a datasheet, a photo or a note and proposes every value that differs from the record. The proposals wait in the Review queue: nothing on the record changes until someone accepts them.

Write a document by voice, or from a photo

In the mobile app, Ask ARGUS → + offers two ways to start a document without typing it:

• Write a document by voice: record what you would write. The workspace’s speech-to-text model turns it into text (AI endpoint → Speech model); nothing of the recording is kept by the transcription.
• Write a document from a photo: photograph a page, a sign or a whiteboard. Its text is read on the phone; the photo is not sent anywhere to be read.

Either way the editor opens with the text, its first words as the title, and a note saying where it came from: check the words before you save. The recording or the photo is kept with the new draft as its source.

Tidy with AI

In the document editor, Tidy with AI writes rough text (a dictation, a photographed page, notes) up as a document: headings, numbered steps, clear wording. It saves nothing: check the result, and Undo puts your own text back.