Using Chat Agent
The Chat Agent is a conversational assistant that lives to the right of the editor, in the file panel. Ask it questions about your script, request help with specific scenes, or seek advice on character development — and unlike a plain chatbot, it can actually do things: it reads your project, searches your Library, and (with your approval) writes Library pages, updates outlines, and drives the app for you.
Solo files and team files
On a file only you can see, the panel is a private conversation with the assistant — just type.
On a shared file, the same panel is the team chat room. Messages go to your collaborators; mention someone with @, everyone with @all, and bring the assistant into the conversation with @ai.
Skills
Type / in the composer to pick from the assistant's built-in skills — prepared multi-step jobs like Build the library, Library gaps, Storyboard a sequence, Coverage, Beat sheet, Character arcs, Dialogue pass, Logline, and Scene notes. A skill runs like any other conversation turn: you can watch it work, interrupt it, and question the result.
Tasks
The Tasks button in the chat panel opens the agent runner: longer preset jobs that run with a live step trail. You can cancel a run, resume it after a reload, review the history of past runs, and jump to the Library page each run wrote.
What it may change
Whether the assistant asks before changing your project is up to you — the Changes to your project setting in Settings → AI offers ask for everything, ask for destructive changes only (the default), or don't ask. A journal records what it changed so the work can be undone, and with AI checkpoints on, a snapshot of the project is saved before destructive changes.
Anything it writes into the Library is attributed to the assistant, so AI-authored content is always distinguishable from yours.
Extra tools from MCP servers
If you have connected external MCP servers, their tools are folded into what the assistant can do, alongside its built-in ones.
Models, cost, and stopping
The chat header has a model switcher, so you can change which model answers mid-conversation without a trip to Settings; the default comes from the Chat row of your task routing. A usage bar in the panel keeps a running tally of tokens and estimated cost.
A Read aloud button on messages speaks a response out loud when a voice is available.
Stop a runaway turn any time with ⌘. / Ctrl+., or the Stop notice that appears while the assistant works.