Agent

On the Agent screen you chat with a single research agent to handle anything from quick questions to serious research work. You don’t pick a different agent for each kind of task: the research agent decides what to search and how to answer based on your request. The screen header shows 연구 에이전트 (“Research Agent”), in Korean even in the English UI.

It’s also the first screen that opens after you sign in.

What can you ask it to do?

Task	Example request
Literature review	“Find and summarize recent papers on solid electrolyte interface stability.”
Reading a paper	“Summarize the key contributions of arxiv.org/abs/2401.12345.”
Working out hypotheses	“Suggest research hypotheses that could explain this observation.”
Critique and review	“Critique the abstract below as a reviewer would.”
Reviewing my own work	“Look at my AI Cortex and tell me this week’s priorities and what’s blocked.”

Where does it look?

The research agent searches on its own when it needs to.

Scholarly literature: arXiv, PubMed, Semantic Scholar, OpenAlex, Crossref, Europe PMC
The web: general web search
Your AI Cortex: only when you choose the Tasks or Web + Tasks scope. See Choose a search scope.

Chat with the agent

In the sidebar on the left, select New chat, or open an earlier chat under Chat history. Chat history is saved automatically for the current workspace.
Type your question in the Ask 연구 에이전트... box and press Enter to send. To add a line break, press Shift+Enter.
The answer appears as it’s written. While the agent searches, you see statuses such as Analyzing the question, Searching the literature… and Searching the web…. When it finds material, Reading sources appears along with the titles of what it found as chips.
To stop an answer, select the Stop response button to the right of the box.

A new chat is titled 연구 에이전트 at first. When you send your first question, the title changes to your question.
Answers are written in the app’s language setting (Personal settings → General → Language).
If you switch to another chat while an answer is being written, generation continues and the answer is saved. If you leave the Agent screen entirely, though, any answer in progress stops.
When you send a question from the box on the Start screen, a new chat is also created on the Agent screen and the answer starts right away.

Check the sources

Answers that use web or scholarly sources show source links throughout the text; select one to open the original in a new tab. Answers that used AI Cortex show the AI Cortex documents they used as chips below the answer; select a chip to go to that document.

When an answer runs long, the agent automatically keeps writing and shows Continuing the response (n). If it still hits the output limit, a note starting with The response may be incomplete… may appear below the answer.

Choose an AI model

In agent chats, you can choose Auto or a model allowed by the current workspace’s plan. Auto is the default. Model selection is separate from search scope: changing models doesn’t change whether the agent searches the web or AI Cortex. Use Context scope to choose that.

Model	Usage multiplier	Free	Plus	Business
Claude Opus 5.5	2×	—	✓	✓
Claude Sonnet 5.5	1×	—	✓	✓
Claude Haiku 5.5	0.1×	✓	✓	✓
GPT-6 Astra	5×	—	—	✓
GPT-6.1 Sol	1×	—	✓	✓
GPT-6 Luna	0.1×	✓	✓	✓
GPT-5.6 Terra	1×	✓	✓	✓
Grok 4.7	0.75×	—	✓	✓
Kimi K3	1.5×	—	✓	✓

How does Auto work?

Auto automatically chooses an appropriate model for your plan. When you choose Auto, the usage multiplier stays fixed at the displayed Auto multiplier.

If a model refuses to respond

With Auto, the agent retries. When you choose a model directly, the request is declined and a reason may be shown.

Model selection applies only to agent chats. The server automatically chooses models for AI Assist, image text extraction (OCR), AI Cortex and other features.

Choose a search scope

Select the button to the left of the message box to open the Context scope menu.

Scope	What it uses	Good for
Web + Tasks	Uses the web together with your AI Cortex	Comparing or extending your work with outside sources
Web	Searches the entire web	Recent literature, general questions
Tasks	Uses only your AI Cortex (auto-organized wiki)	Checking progress, open loops and decisions

New chats start on Web. When you reopen an earlier chat, the scope you last used in that chat comes back.
Tasks and Web + Tasks need AI Cortex turned on in the current workspace. If it’s off when you send, you see Enable your AI Cortex first to use work context.
Only you can see your AI Cortex, so the agent never reads another member’s AI Cortex either.

Review AI Cortex edit proposals

If you ask for a change to your records in the Tasks context, such as “Change my current focus,” the agent can create an AI Cortex edit proposal. The agent never changes your personal records directly.

In the notice An AI Cortex update proposal has been created. Review it before applying., select Review to go to the AI Cortex screen. Compare the proposal with the current content, then apply or reject it. Applying creates a new document version. For the full flow, see AI edits in AI Cortex.

Copy answers and save them as idea cards

Use the copy button below an answer to copy its text. Open the More (⋯) menu to see when the answer was written and to keep it as an idea card.

Create an idea card from a chat

Open the More (⋯) menu below the answer you want to keep.
Select Save as idea card.
If you belong to one workspace, it’s saved right away. If you belong to several, choose a workspace under Where should this be saved?.

The card is created on that workspace’s default board, and sources used in the answer are linked to the card as references where possible.
When it’s saved, Idea card created. appears with View on board.
If the answer proposed hypotheses, the Create idea card window opens. Choose the hypothesis and board, refine the title, content, tags and references to attach, then select Create card.

Guests can’t save idea cards. You’ll find saved cards on your idea board.

New chats are visible only to you by default. The person who created a chat can open its ⋯ menu in the chat list on the left and select Share to workspace to make it visible in the current workspace. Select Unshare in the same menu to make it private again.

Shared chats show a person icon (Shared to workspace), and chats created by someone else show the creator’s name.
Only the chat’s creator can share, unshare or delete it. Admins can’t share or delete someone else’s chat on their behalf.
Members can open a shared chat and keep the conversation going. Usage keeps counting toward the workspace the chat belongs to.
Guests can only read shared chats (Guests can view shared conversations in read-only mode.). They can’t start chats, send messages, share or delete chats, or save answers as idea cards.

When a chat is shared or unshared, the relevant members get an Inbox notification.

Delete a chat

In the ⋯ menu of a chat you created, select Delete to delete it immediately, without a confirmation step. There’s no way to bring back a deleted chat, so be careful.

Usage

The agent shares the workspace’s AI credits with AI Assist and AI Cortex. When the monthly credits run out you see You've used all of this month's AI credits. They reset on the 1st of each month. On Plus and Business, credits grow with purchased seats and are shared by members. For the same seat count, Business includes more credits. See Plans and limits.