The Lavori AI assistant

The assistant is available everywhere as a slide-over panel and as a full-screen workspace. On desktop the panel opens inline alongside whatever page you are on; on mobile it slides in as an overlay. Either way it is labeled Lavori Assistant and opens with a focused text box plus a few suggested prompts.

The dedicated AI assistant page gives you the same chat with a saved-conversation sidebar. You can search past chats by title or message content, switch between them, start a new chat, and delete old ones. Both surfaces talk to the same backend, so a question works identically whether you ask it from the panel or the full page.

• Slide-over panel — quick questions without leaving your current page.
• AI assistant page — full chat history, search, and conversation management.

When you open the panel, the suggested prompts change to match where you are. On the projects area it offers things like asking which projects are behind or creating a new project; on scheduling it suggests checking who is scheduled this week or which shifts are unfilled; on billing it suggests recent invoices or outstanding balances.

The assistant also receives context about the page you are viewing — the entity type, its ID, and the active tab or section. That means when you say 'this project', 'this invoice', or 'that task', the assistant resolves the reference to the item on screen instead of guessing. It also tracks references back through the conversation, so follow-ups like 'now mark it as done' apply to the thing you just acted on.

The assistant reads your organization's data through a set of search tools and answers in plain language. It can look across projects and tasks, the contact and team directory, the schedule of events and shifts, invoices and quotes, budgets and the Service Library of billable items, and project note pages.

It can also pull aggregated analytics — totals and trends such as revenue, outstanding balances, crew utilization, overdue tasks, project summaries, and budget variance — for a date range or a single project. If a search returns nothing, it tells you rather than inventing an answer.

• Projects, tasks, sections, and statuses.
• Events, shifts, and open or unfilled crew slots.
• Contacts, companies, and team members.
• Invoices, quotes, budgets, and Service Library items.
• Project note pages and aggregated analytics.

For admins and members, the assistant can also make changes through scoped action tools — every action is a normal, audited API call, the same one the interface would make. From a single message it can create projects with sections and columns, generate tasks, create and assign shifts, create events, add and update contacts, draft invoices, quotes, and budgets, manage Service Library items, write or edit project notes, and transition statuses such as completing a project or publishing shifts.

Some actions are deliberately gated. Sending a direct message and deleting a note page or billable item require an explicit confirmation first — the assistant tells you exactly what it will do and waits for your yes. There is no general delete tool for projects, tasks, invoices, and similar records; for those it gives you a direct link to delete the item yourself. Admin settings, user and organization management, and file management are outside the assistant's scope and it points you to the right settings page instead.

To keep actions predictable, the assistant works within a few caps. It runs a bounded number of tool-call steps per message — up to five for a simple request and more for a complex one — so if you ask for many actions at once it completes what it can and tells you what remains for a follow-up. Bulk task generation and shift creation are capped at 25 items per call. Questions are limited to 2,000 characters.

On plans that include speech-to-text, a microphone button appears in the chat input. Record a question by voice and the assistant transcribes it and submits it for you. If no speech is detected it lets you know so you can try again. Recordings are limited to common audio formats and a 25 MB size.

Replies stream in token by token rather than appearing all at once, so you see the answer take shape immediately. While the assistant is running a tool — for example searching your projects or creating a record — the panel shows which action is in progress, and the connection is kept alive during those gaps. You can stop a response mid-stream at any point.

Conversations are multi-turn: the assistant remembers earlier messages in the same chat and uses them as the source of truth for anything it just listed, created, or modified. You can keep up to 25 saved conversations per organization.

AI availability depends on your plan. If your plan does not include AI, the assistant shows an explanation and a link to upgrade. Plans that include AI come with a monthly request allowance; as you approach it your organization is notified at 80%, 95%, and 100% of the quota. When the limit is reached, depending on your plan's overage policy the assistant either keeps working or pauses with a clear message until the next billing cycle resets the quota.

Everything is scoped to your role and your data. Contractor accounts get a read-only assistant focused on their own work — their assigned tasks and shifts, the projects they belong to, and those projects' structure. Contractors do not have access to billing, budgets, reporting, the contacts directory, or any action tools. The assistant never reaches data outside your organization.