Chat

Talking to the agent

You talk to the agent in the Right Panel. Get suggestions for plot or lore, discuss where to go next, and work through the places you are stuck. The agent knows the screen you are on and the context of the Space, so its answers fit your work. When it needs to, it finds and reads the Manuscript and the World, and hands jobs like drafting, Revision, and Proofreading to subagents. Characters, places, and lore settled in conversation are filed into the World as nodes when you run Organize World. The World starts empty, and the agent creates the types it needs when it organizes. (Organize World)

Ask the agent to develop a character — background, personality, motivation, relationships — and it proposes what goes into that character's World node. With Edit permission on Ask first, you approve it before anything is written. You can also ask how TypeTak itself works — “Can TypeTak do this?”, “How do I use the World?” The agent looks it up in this guide with Search service guide (typetak_info_search) and answers from what it finds.

What the agent draws on

The agent answers and writes from what is already in your Space. Most of it builds up as you write and talk, so you don't restate your lore with every request.

• The World: it searches and reads the nodes a request needs (world_find, world_read).
• The World Setting: what holds for the whole work — its era, its stage, its social order — read when the writing at hand leans on that frame (system_doc_read).
• The Writing Style: the style set for each manuscript, used in drafting and Revision.
• Guidance: how you want the agent to work.
• Pages: the page you are on, any page you mention with “@”, and other pages it finds and reads itself (page_find, page_read).
• The web: facts from outside your work (web_search).
• World Organizing Rules: the rules of this Space the agent follows when it writes into the World with Organize World.

Choosing a model

Pick the AI that runs the conversation.

• Model: Choose GPT or Gemini in Agent settings, the slider button in the input box.
• Detailed models: On the Max plan, you assign a model and a Reasoning level per feature in Settings > Detailed models. Under Models by feature the rows are Agent, Revision, Canvas, Feedback, Proofreading, Enhance, and Rephrase, and under Models by draft stage are Draft, Self-evaluation, and Final.
  • Each row leaves the model unset or takes its own — GPT, Gemini, or Claude. An unset row runs on the model TypeTak picked for that feature. Once a row has its own model, you also pick a Reasoning level: Low, Medium, High, Very high, or Max. While the model is unset, the Reasoning level stays locked.
  • One Detailed models switch turns the whole set on or off. The same switch is in both Agent settings and Settings > Detailed models, and while it is off, the rows are not applied. While it is on and any row has its own model, Model in Agent settings is grayed out.
  • On other plans the rows are visible but locked, with a note that they can be changed on the Max plan.

The language the agent answers in is set under Preferred response language in Settings > Agent. (See Customization.)

Adding context

• @Page Mention: Type “@” in the input box to point the agent at a specific page as context. The page you mention shows as a chip in the input box.
• Attach file: Attach a file to the chat. For each file, decide whether to save it into the Manuscript or Documents, or use it only as reference for that one request.
• Web search: When your work needs a fact from outside it, the agent searches the web, shows what it found and the Sources it used, and uses it in its answer.

Seeing what it is doing

While the agent works, each step it takes appears in the chat as a short line — Read page, Find in world, Planning the draft, and so on — with a pulsing dot while that step runs. Once a step is done, press the page or World node it names to open it. The draft, Revision, and Organize World cards each show their progress. While the agent is thinking, Thinking appears.

Keeping going while it answers

• Stop: The stop button (■) halts the reply.
• Queue: A message you send while a reply is streaming shows dimmed above the input box, and is sent as a new message once the current reply ends. Several messages are merged and sent as one. Send it with the round button to the right of the stop button, or press Enter.
• Do something else: The agent keeps working on your request after you leave the conversation. Move to another conversation or keep writing in the editor, and you get a notification when the job finishes.

When the agent asks you

When the agent needs a decision only you can make — an unclear request, which direction to take, how far a change should go — it stops and asks. The question appears in place of the input box. Pick an answer (some questions take more than one), or write your own in Other answer. With several questions, Next question moves on, and Send on the last one sends all your answers. The agent then carries on.

Rating and copying replies

Each finished reply has three buttons underneath: Rate response, Copy, and Branch from here.

• Rate response opens Helpful and Not helpful. Not helpful asks you to write why the reply missed, and that reason goes to TypeTak with your rating.
• Once you have rated a reply, the same button reads Remove rating and takes the rating back in one press.
• Copy copies the whole reply.

Branch

Branch off mid-conversation to try another direction. Press Branch from here next to Copy on a finished reply, and the conversation splits right at that point. You can go back to the earlier conversation at any time, so you compare several directions without losing the thread.

Starting a new conversation, and conversation history

Start a new conversation when the topic changes. Earlier conversations stay on record, and you can reopen them whenever you want or delete the ones you no longer need.