Working with the assistant

The assistant sits beside every page in Twin, answers questions and does the work on your projects, clients, tasks and documents.

Where the assistant lives

The assistant is a permanent column on the right of every page, so there is nothing to open. A new conversation shows What can I help you with? and three starter chips (Add new client, Ask an EIC question, Create new project) that fill the message box for you to finish. Drag the column's left edge to widen it, or click Fold panel to collapse it into a narrow rail.

Conversations, tabs and history

Every conversation gets a numbered tab. A fresh tab reads New; after your first message Twin gives the conversation a generated title, so there is nothing to rename. Close a tab with the x next to its title. Nine tabs fit; opening a tenth closes the one you have not used for longest, which stays in your history.

The list icon at the far left of the tab strip opens All chats: a searchable history grouped under This week, This month and Older. On a project page, the dropdown beside the search box narrows it from All chats to This project.

Sending a message

1. Type in the box at the bottom of the column and press Enter. Shift+Enter starts a new line.
2. Click Add files, or drop a file onto the box, to hand something over. What you attach becomes a source on the project, available to everyone and to every document in it. See Sources: what Twin writes from.
3. While it works, the send button becomes Stop and a status line names the current step. When it is done, those steps collapse into a line such as Reasoned for 6s that you can click to expand.

What it can do

• Create and change projects, clients, tasks, documents and sources.
• Know which page you are on, and offer an Open button that takes you to the right place.
• Search the grant database for a scheme's provider, deadline, target groups and amounts. See Which grant schemes Twin supports.
• Do calculations, and research the public web once you allow it.

When it asks permission

Two things need your approval, and the assistant asks for them in the conversation itself.

• Allow web search? the first time a conversation wants the public web. Once allowed, it stays on for that conversation and a Web search chip sits above the message box.
• Grant this conversation access? when it wants to edit a further project, beyond the one the conversation is already working on. Choose Grant access or Cancel. The outcome is written back into the conversation, and a decline is treated as a decision, not a retry.

How it behaves

• It says what it can and cannot do rather than promising work it will not deliver. Ask it directly if you are unsure.
• It drafts a project plan and waits for your approval on the project page before writing sections. See Projects: where an application comes together.
• When the plan changes, it re-checks the sections already written against it.
• If something it started in the background fails, it reports the failure in the conversation instead of dropping it.
• What it learns about your firm, a client or a document type is kept as memory. See Memory: what Twin remembers.

Long conversations

A conversation carries everything said in it, so it fills up. Well before that, a warning appears above the message box: This conversation is getting long and will soon reach its context limit. Start a new conversation to continue smoothly. At the limit the message box is replaced by a short notice and a Start new conversation button. Nothing is lost: the old conversation stays in All chats, and everything the assistant changed is on your projects and documents.

Good to know: one conversation per subject beats one long thread, and each conversation keeps its own web-search and project permissions.