The chat assistant
Steer your shadow-planner plan by chat — @-mention projects and people, use slash commands, and approve or decline every risky change before it happens.
The chat drawer sits in the bottom-right corner of the window (drag its edges to resize it). It is scoped to the active scenario: each scenario keeps its own conversation, and the header always tells you which provider and model you're talking to.
You can ask questions ("who is overloaded in September?"), or make changes in plain language ("move the design review to next Friday and shift whatever depends on it"). The assistant reads and writes the same plan you see — every change shows up immediately in the Gantt and lists.
Referencing things precisely: @-mentions
Type @ to open an autocomplete of your projects, people, labels, epics, milestones and work items, and pick one to insert it as a chip. That way the assistant operates on exactly the entity you meant instead of guessing by name. Pasting text that contains known names converts them into chips automatically.
Slash commands
Messages starting with / are instant, deterministic commands — they never go to the AI model:
| Command | What it does |
|---|---|
/status |
Upcoming milestones across all projects — a 60-day lookahead with on-track / at-risk / behind / overdue chips |
/export |
Downloads the current chat history as a CSV file |
/new |
Starts a fresh conversation (same as the New chat button) |
Nothing risky happens without your approval
The assistant applies simple creates and updates directly — each one appears in the transcript with a WRITE badge so database changes are always visible. Three kinds of operations pause for your explicit Approve / Cancel:
- Deletes — deleting a task, epic, project, person or label shows a cascade preview first: how many epics, tasks, assignments and tags would go with it.
- Cascading date changes — if moving a task would shift dependent tasks or auto-extend an epic's or project's date range, a "Cascading side-effects detected" card lists every knock-on effect. Declining cancels the original change too — nothing is ever applied halfway.
- Bulk changes — assigning several people at once or tagging many items shows the exact list of affected entities before anything happens.
Everything the assistant does is also recorded in the audit log with your prompt and its reasoning, and its changes are undoable just like your own.
Conversations
- New chat starts a fresh conversation; the old transcript is kept on disk but not shown.
- The trash icon permanently deletes the current conversation after a confirmation.
- Conversations survive app restarts, and switching scenarios switches to that scenario's own chat.
Behavior worth knowing
- The assistant shows one page of results at a time and asks before fetching more — it won't dump your whole database into the chat.
- When a request is ambiguous, it asks — sometimes with a clickable choice list.
- When a change auto-extends a parent project or epic, a toast tells you exactly what moved (for example "Auto-extended Project 'Mobile': end_date 2026-08-31 → 2026-09-15").
Which model answers, and how to switch it, is covered in Connect an AI provider.