Plan a project with AI

The planning agent turns a written brief into a structured plan proposal: a project with epics, tasks, estimates and assignments. Crucially, it is a draft — nothing is written to your plan until you have reviewed it and pressed apply.

Starting a plan

Use one of the Plan with AI buttons in the Gantt or Projects view (or the entry in the settings gear menu). The buttons are greyed out with "Set up an LLM first" until an AI provider is connected — they activate the moment one is.

You can plan a new project or add to an existing one, and drafts can be saved and resumed later, so a half-finished planning session is never lost.

The planning wizard

1. Input. Choose between a One project or Multiple projects template, optionally set the project name (a name you type here wins over whatever the AI suggests), and write your brief in Markdown or plain text. You can also import a document — .docx, .txt or .pdf up to 5 MB — and it is converted to Markdown to pre-fill the brief. PDF extraction is best-effort: scanned or multi-column PDFs often come out garbled, so convert those to .docx or plain text first.
2. Analyze. The AI streams its draft while a live Plan structure preview builds up as a tree — project, epics, tasks, with type chips and estimated person-days. The raw model output is available behind a collapsed Raw output disclosure, and a meta line shows which provider and model produced the plan plus token usage.
3. Review. The draft is rendered as an editable Markdown document: a # Project: heading, ## Epic: sections, and one - [ ] line per task with optional [duration, assignees] metadata such as - [ ] Wireframes [3d, Alice]. Your edits win — whatever the document says after you edit it is exactly what gets applied.
4. Apply. Materialise the plan. On success the app navigates to the new project.

What materialising does

• Creates the project, one epic per section and one task per checklist line — all-or-nothing in a single transaction. Either the whole plan lands, or nothing does.
• Resolves assignee names against the people in your scenario by fuzzy name match ("Alice" finds "Alice Weber").
• Validates dates and auto-extends parent epic and project date ranges so they always enclose their children.

Good to know

• Materialising counts as a bulk import, so it is not undoable with Ctrl+Z — if you want a safety net, take a snapshot first. The draft itself stays available either way.
• The planning agent uses the same provider you picked for chat. Long briefs on cloud providers cost accordingly; on Ollama everything is free.
• If the plan is too large for the model's output budget, the error tells you exactly which setting to raise — or shorten the brief and plan in stages.

For a full worked example, read the walkthrough: Planning a project with a local LLM.