The AI assistant

Click AI Chat in the toolbar. A panel slides in from the right, headed AI Assistant with a Beta badge. On a wider screen the panel sits beside the chart, so you can keep working on your plan while it is open. On a phone it fills the screen.

The line under the heading names your level and the model in brackets. It reads Standard on Team Member and Project Coordinator, and Premium on Project Manager. All plans use the same chat model. They differ in the monthly allowance, described under AI usage and limits, and in what the assistant offers on its own, described under what you can ask.

Opening the panel collapses the project sidebar to give the chart and the chat more room. The first time this happens in a browser, a Sidebar collapsed message points to the panel icon at the top left of the toolbar, which brings the sidebar back. Press Esc or click the close button to hide the panel. Your conversation is kept while the panel is closed, so reopening it picks up where you left off.

A fresh conversation opens with "How can I help?" and four suggested prompts matched to your role, such as a review of the project against PRINCE2 or a summary of the biggest risks. Clicking a prompt runs it immediately. You can also type your own message and press Enter to send it (Shift plus Enter adds a new line). The assistant reads the first 5,000 characters of a message and ignores the rest. The New chat button at the top clears the conversation, and switching projects starts a fresh one automatically.

Replies can contain formatted lists, tables, and links. A link in a reply opens in a new tab.

The assistant works on the project you have open. With no project open, use the Build with AI choice on the start screen instead, which opens the AI Project Wizard.

The assistant needs a signed in account. Guests see a sign in prompt instead of the message box; see trying Ganttastic without an account for what guest mode does include.

In everyday terms, the assistant can:

• Add, update, and delete tasks (dates, duration, progress, notes, and predecessors).
• Add milestones. A milestone the assistant adds always has a duration of 0 days.
• Add dependencies between tasks.
• Add a sub-task under a phase, or split a task into parts. When the task sits in a phase, the parts stay in that phase. Splitting a task that is not in a phase puts the parts under it as sub-tasks.
• Make project level updates.
• Carry out plans of up to 15 steps in one turn, executed in order. Bigger requests are split into batches, and the assistant confirms with you before continuing.
• Run read-only what-if simulations.
• Ask short clarifying questions with clickable options when your request is ambiguous. You can click an option, type a free text answer, or tell it to skip and use its best guess. After two rounds of questions in a row, it stops asking, says it has enough to go on, and proceeds with sensible defaults you can change later.
• Open the AI Project Wizard when you ask it to create a new project. Your message becomes the brief word for word (up to 3,000 characters), with a project name taken from your message when it has one.

Example prompts that work well:

• "Add a task called Budget approval starting 1 June, lasting 5 days."
• "Which tasks are overdue?"
• "Summarize the biggest risks on this project in PRINCE2 terms."
• "What happens if task 2.4 slips by a week?"
• "Who is assigned to Order the signage?"

When you ask a short, open question about a single task, such as "explain 3.4", the assistant first asks how much detail you want: Explain in easier terms, Break down what I need to do, Deep dive or Just the basics. A deep dive uses more AI tokens. If your question names what you want to know, such as the start date or the predecessors, it answers straight away.

To keep each request small, the assistant reads the notes on phases and on tasks outside a phase only when your message uses the word note, notes, description, detail, comment, or context. Notes on sub-tasks are always included. It reads your project’s resources and who is assigned to each task only when your message uses a word such as who, team, assigned, resource, person, or owner, or names one of the project’s resources in full.

Anyone you share a project with can open the assistant on it. What they can do depends on their role on that project:

• A Viewer can ask questions and run what-if simulations. Any change a Viewer asks for is declined with "You have read-only access to this project. I can answer questions about it, but I can’t add, edit, or remove tasks. Ask the project owner to give you Editor access."
• An Editor can ask for changes to tasks and accept them, like the owner.

Tokens are charged to the person who sends the message, not to the project owner. A declined request still counts any tokens the AI used to read it. See share and collaborate for roles.

Any action that would change your project is held in a preview panel. Nothing runs until you approve it.

• The line above the card says what the proposal does, in words built from the proposal and your edits on the card, so it matches what Accept applies. The assistant’s own explanation sits under Show the assistant’s notes; open it to read why it suggested the change. The notes start open when they say the plan is only a first batch, or when the line cannot name a task.
• For a single action the button reads Accept. For a plan of several steps it reads Accept All, or shows a count (for example Accept 3) when you have skipped some steps. Reject discards the whole proposal.
• Each step in a plan has its own include or skip toggle. Unticking a step also unticks any later step that needs it, such as a link to a task the unticked step would have created. Those steps read Skipped: it needs a step you removed. Ticking a skipped step again also ticks the steps it needs.
• You can click a proposed value to edit it before accepting. A step you changed is tagged edited.
• Fields the AI estimated rather than took from your message are outlined in amber and tagged AI estimate, so you can confirm or correct them. If a proposed task has no name, accepting is blocked until you type one.
• When the assistant places a new task under a phase or another task, the card shows Placement: Sub-task under and that task’s name, with a link to make it a top level task instead. Once moved, a second link puts it back.
• A new task has a Set as milestone link that makes it a milestone of 0 days, and a milestone has Remove milestone, which makes it a task of 1 day.
• A new task that will be linked to another one shows Depends on and the task names.
• Proposed dates that land on a weekend or a holiday are adjusted to your work week automatically, with a visible note of the adjustment.
• A date you type is checked as you go. A date the app cannot read shows Invalid date. Use YYYY-MM-DD. A date on a day off shows Non-working day with a snap link that moves it to the next working day.
• On a proposed update, a duration or progress value outside what a task allows shows a note with the value that will be saved instead, for example 100 for progress above 100.

While a proposal waits, its card is headed Proposed (or the number of proposed changes for a plan). Once you decide, a small tag under it records what happened: Applied once it ran, Not applied when you cancelled a delete, and Not completed when it failed. A rejected proposal shows no tag, and a what-if always reads Simulated.

The chat then confirms the result in its own words:

• "Changes applied successfully." for a single change.
• "All 5 steps applied successfully." for a plan you accepted in full.
• "Executed 3 of 5 steps. 2 steps were skipped." when you unticked some steps.
• A sentence starting "With your edits:" that lists the values you changed on the card, because the assistant’s own reply cannot know them.
• "Changes discarded." after Reject.

A plan runs its steps in order and stops at the first step that fails. The steps before it stay saved, and the chat says so, for example "Some changes could not be applied. Executed 3 of 5 steps, then step 4 failed and the plan stopped. Use Undo to restore your project."

A single delete proposed by the assistant opens the same delete confirmation as deleting a task yourself. If you cancel it, the chat says "The delete was cancelled. Nothing was changed." A delete that is one step of a larger plan runs when you accept the plan.

Every applied action gets an Undo button in the chat that restores the project to its state before that action, including after a plan that stopped part way. Undo works only while the project still looks the way the AI left it. Once you or someone else has changed the project since, it answers "Undo is no longer available. The project has changed since this message." In that case undo your later edits first, or fix the tasks by hand. Ctrl+Z (Cmd+Z on Mac) undoes as usual, and Ctrl+Y or Cmd+Shift+Z redoes.

• Assign owners to tasks. Chat actions carry no assignment field, so a request like "assign it to Sam" cannot be applied from chat. Ask the assistant to add the task, then set the owner yourself while working with tasks. The AI Project Wizard is the one AI surface that suggests owners.
• Compare progress against your baseline. Baseline data is not sent to the chat model. Use the tools described in baselines and variance instead.
• Make very large structural changes in one step: a re-baseline of the whole plan, putting the whole plan or project in a new order, restructuring it, or moving a whole phase with all its sub-tasks. The assistant replies that it is a large restructure and asks for the first concrete change, then works through it with you step by step.

All AI features in Ganttastic share one monthly token allowance per user. Both what you send and what the AI writes back count toward it.

Allowances reset on your billing anniversary. Free accounts reset on the day of the month the account was created.

The sidebar footer shows a compact AI meter on every signed in account: a percent used token bar, how many AI plans you have used this month, and a reset label such as "resets in 5 days". Inside the chat panel, a Usage bar sits in the header and a running token counter (used and limit) sits next to the message box. The meter can legitimately show more than 100%, because a request that starts under the cap is allowed to finish.

Before a message goes to the AI, Ganttastic adds an estimate of its cost to what you have used. If the total would pass your allowance, the message is not processed, so a long message can be refused a little before the meter reads 100%. Instead you get an upgrade card that says you have reached your monthly AI capacity and gives the date your usage resets. Starting a new wizard generation is blocked earlier, from 90% of the allowance.

Two kinds of request are handled by Ganttastic without the AI model, so they cost no tokens and still work when you are at your cap. The first is a message that starts by asking to create, build, or make a project, such as "create a project called Office move", which opens the AI Project Wizard. The wizard itself still blocks a new AI plan from 90% of your allowance. The second is adding a task by name alone with no dates, such as "add a task called Budget review". That task arrives in the preview starting today (or the next working day) and lasting 1 day, both marked as AI estimates for you to adjust.

The upgrade card names the next plan with its monthly tokens and price. On Team Member and Project Coordinator it offers Project Manager with Upgrade now. On Project Manager it offers Enterprise with Enquire about Enterprise, which opens the enquiry form in a new tab. See all plans opens the pricing page.

Separately from your own allowance, Ganttastic has one shared daily AI limit for all accounts together: 5,000,000 tokens a day, counted on Brisbane time. If it is ever reached, the chat shows "Ganttastic’s AI Assistant has reached today’s shared usage limit. It resets overnight (Brisbane time). Please try again tomorrow." and the AI Project Wizard stops too. Upgrading does not help in this case; AI comes back the next day.

While the assistant works, the chat shows what it is doing (Reading your plan, then Working out the answer, then Writing the reply) with a running timer. After 10 seconds it also shows short tips about Ganttastic. After 30 seconds it adds "Still working. Bigger changes can take up to two minutes." After 2 minutes with no reply, the chat stops waiting and shows No reply yet. The request can still finish on our side and count toward your AI usage.

Problems appear as a block above the message box, with a Retry button that sends the same message again (the chat adds "Sending that again…") and a close button to dismiss it. You do not need to retype anything. Messages you can see there include:

• No internet connection: reconnect, then press Retry.
• "Something went wrong on our side and the assistant could not answer. Nothing in your project changed."
• A message that the AI service is busy, unreachable, or temporarily unavailable: wait a moment, then press Retry.
• "Your account is still being set up. Try again in a moment.": this appears for a few moments after you create an account.

If you are offline when you press send, nothing is sent and your message stays in the box until you reconnect.

If Ganttastic reloads the page by itself to recover from a sync problem, the conversation comes back with a line saying it was restored. A proposal you had not accepted still waits for Accept, and Undo is not available for changes made before the reload.