NavantoAI

NavantoAI is the assistant built into Navanto. On the NavantoAI Chat page you can ask about anything in your organisation’s data, projects, tasks, contracts and the rest, and have it do work for you; it reads what is actually in Navanto now, not a copy. It is for your team; clients do not have it.

Chatting

Ask in plain words and the answer streams back. You can attach any file, up to 10 per message, and the assistant can hand files back to you as downloads, and show photos from your projects.

Before it changes anything, it asks: it describes exactly what it is about to save, send or delete and waits for your agreement, and telling it to go ahead in your message counts as that agreement.

Clear chat wipes the conversation with no way back: on the web it goes immediately, with no confirmation; in the iOS app it asks first. Copy anything you need before you clear.

On the web a conversation stays with the browser tab you started it in; in the iOS app it picks up the same conversation each time you open the tab, until you clear it. Either way, after half an hour idle the assistant suggests a fresh chat; long conversations drift, and a new one usually answers better.

In a thread

Mention @NavantoAI in a project post and it replies there, in the thread, the way a colleague would: it reads what has been said and looks at the project with the same access your role gives you. Ask it to do something — create a task, move a date, log a note — and it does it. The mention is the go-ahead, so unlike the chat page it does not describe the change and wait for your agreement first; it does exactly what you asked, nothing wider, and its reply says what it changed. If your request could mean two things, it asks in the thread instead of guessing. Whatever your role cannot do, it cannot do either, including anything that reaches the client, and every record it changes shows up in the activity feed as NavantoAI. Once it has joined a thread you can carry on talking to it without mentioning it again: it reads each reply and answers when the reply is for it, and stays out of it when you are talking to a colleague.

What it remembers

The assistant keeps a memory for your organisation: your conventions, preferences and standing context, shared across everyone’s conversations and with NavantoAI in project threads, so something it learns in a thread it knows in chat, and the other way round. Under Settings, on the Organisation page, Show summary writes you a summary of what it currently holds.

To correct it, tell the assistant in chat; it updates its own memory. Clear memory is for administrators only and cannot be undone.

Where else AI appears

• Risk briefing on a project is written every weeknight for every live project and rewritten within the hour when someone marks one of its findings actioned, reading across the whole job and settling on a verdict — on track, watch, at risk or critical — that the project’s badge shows. Ask about a project’s risk in chat or in a thread and the answer comes from that briefing, and ask which of your projects are at risk and it reads every project’s verdict at once; it can also tell you what is new or changed since the last one, how long a finding has been open, and why something isn’t being flagged, including who dismissed it and the reason they gave. Tell it a finding doesn’t need raising and why, and it will dismiss it for you, passing your reason on in your words; tell it you’ve dealt with one, and it will mark it actioned with what you did, so the project is rechecked within the hour. In chat it checks with you first; in a thread, asking is enough. It only ever dismisses or marks actioned what you ask it to.
• Analyse with AI on a drawing revision fills in details read from the sheet.
• Week summaries appear over each week of site updates, generated on a schedule.
• Generate description drafts a client update from the week’s site updates, for you to edit before sending.
• Discuss a price starts a new chat that works out a variation’s price with you against your catalogue and contract, saving the agreed figure to the draft.

Check its work

AI can make mistakes, and Navanto says so wherever it appears. Read anything it produces the way you would a colleague’s first draft, especially anything going to a client or touching money.