Chat

Through chat, you can converse with various LLM models and utilize tools to automate your work.

Feature Guide

Model Selection and Option Settings

You can change the LLM model used for conversation through the model selector at the top of the chat screen. Each model provides various options tailored to its purpose. You can select different models for each browser tab, allowing you to compare multiple models simultaneously or use them for different purposes. You can also start a conversation by calling a preset by pressing the @ button in the input field at the bottom of the screen.

Supported Models:

Auto Router

When you select Auto Router, the model is selected automatically based on the request type and conversation context. Use it when you are not sure which model to choose, or when the required model level may vary from request to request.

Auto Router considers both the current request and previous conversation context. For example, short standalone requests can be handled by a lighter model, while follow-up conversations, code or data analysis, file-based requests, and interpretation of tool results can be routed to a higher-tier model.

In the same conversation, the selected model may be reused for a period of time. If a new request includes signals that require a fresh decision, such as code, data, files, or an explicit request for a stronger model, the model is selected again.

Note: An administrator must enable Auto Router in LLM Models. The models it routes to depend on the organization’s configuration.

Reasoning Models

• You can obtain deeper answers by increasing the reasoning intensity when complex problem solving is required.
• You can easily adjust the reasoning level through the slider.
• When using Reasoning Level, an option to expand and view the AI’s thinking process is provided. Click the thinking process area in the response to see the logic the AI used to derive its answer.

Context Expansion

You can enable Wide Context with supported models such as Claude Opus 5.5, GPT-5.4, Gemini 3.1 Pro, and Grok 4.5. See the Wide Context column in the table above for the full list.

• The context window expands up to 1M tokens.
• Useful when you need to analyze very long documents or large amounts of code.
• Wide Context models consume more credits, so please use them appropriately.

LLM Guide for Enterprise Environments in Japan

LLMs that can process data in Japanese regions are available to meet enterprise requirements for data processing location and compliance. These models are provided through AWS Bedrock and Vertex AI, which offer data protection policies such as Zero Data Retention (ZDR) and Data Residency. Review the table below to select a model that meets your organization’s security policy and data governance standards.

Data Residency?

Data Residency restricts where data is stored and processed to a designated geographic area. Models that support Japan Data Residency process your input and output data within designated Japanese regions. Check the table below for each model’s Japan Data Residency support and processing route.

ZDR (Zero Data Retention)?

ZDR means that the model provider does not retain user data in persistent storage after processing a request. For supported models, prompts and generated responses are not retained by the model provider after processing, reducing the risk of sensitive information being stored externally.

Frequently Asked Questions

Q. Why are models provided through API providers such as Amazon Bedrock or Google Vertex AI?

Although using a model provider’s API directly also guarantees ZDR, models are provided through cloud AI platforms such as Amazon Bedrock and Google Vertex AI to offer options that support both ZDR and data processing requirements in Japan.

However, a model provided through Amazon Bedrock may not guarantee Japan Data Residency. Using Amazon Bedrock itself can be an important requirement under an organization’s security and governance standards. Models routed through Bedrock are therefore also available regardless of whether they support Japan Data Residency.

Q. What is the difference between (JP) and (Bedrock) in model names?

• (JP): Models provided through Amazon Bedrock or Google Vertex AI that support Japan Data Residency.
• (Bedrock): Models provided through Amazon Bedrock that do not support Japan Data Residency.

Q. Why does Price Premium apply to some models?

For some models, the provider charges a higher rate than the base model when they are offered in specific regions or with Data Residency options. Price Premium is the percentage of additional credits charged to reflect this difference in official pricing.

For example, a model with a 10% Price Premium consumes 10% more credits than the base credit amount for the same usage. If the base amount is 100 credits, the total charge is 110 credits.

Additional Features

Click the + button on the left side of the chat input field to select file, Skill, generation hint, search, and voice response options.

Items available from the + menu:

• Add photos and files: Upload files in various formats such as documents (PDF, Excel, PPTX, DOCX), images, and more to analyze contents or request summaries. You can upload directly from your local computer or import files saved in My Drive. Document files are processed using Skill-based processing for faster and more accurate analysis. Files are supported up to 200 MB per turn. There is no cumulative limit per session.
• Select Skill: Select a Skill directly in the chat screen or type slash (/) to call the Skill you need. If the Skill you need is not in your Skill list, click Skill management to add it from the organization library and use it right away. See Skills for details.
• Generate image: Generate AI images with text prompts or edit generated images.
• Generate widget: Create widgets that visualize data analysis results as charts, graphs, and more.
• Generate code artifact: Generate artifacts containing executable code and preview screens.
• Create web app: Build and deploy web applications from a conversation.
• Web Search: Activate real-time web search functionality to receive answers based on the latest information.
• AI Voice Reply: Listen to the AI’s response in voice.

Use the Use MCP button around the input field to configure and use additional MCP tools needed for the current conversation without a Preset.

Under the MCP settings list, use Tool Access to choose how tools are loaded.

• Load tools when needed: Finds and loads only the tools that match the question. This is useful when many tools are available because it reduces Context usage and helps select the required tool more accurately.
• Tools already loaded: Loads all tools from the selected MCPs from the start of the conversation. This is suitable when there are only a few tools or when you always use the same tools.

Note: In Agent chat, only add-ons allowed in the agent settings are displayed. Code artifacts and Create web app may not be supported in the mobile app.

Tip: Click the microphone icon on the right of the chat input field to input questions by voice.

OCR and Image Analysis

Images can be analyzed for their contents, charts, tables, and diagrams with models that support image input. For PDFs, text is extracted with OCR and visual materials such as charts and diagrams can also be analyzed.

• Extract text from images and scanned documents to summarize content or find needed information.
• Explain figures, trends, and comparisons in charts and tables from images or PDFs.
• Explain components, relationships, and process flows in diagrams from images or PDFs.

Note: With Auto Router, only OCR-based text recognition is available. Visual analysis of images, charts, and diagrams is limited.

Examples:

• “Summarize the notable trends and causes in this attached sales chart.”
• “Explain the components and data flow in this architecture diagram.”
• “Find the contract term and termination conditions in this scanned contract image.”

Conversation Convenience Features

Tool Execution Description

When a preset uses a tool, easy-to-understand descriptions like “Checking weather…” are displayed. Tool call purpose descriptions are automatically displayed according to the user’s language settings (Korean, English, Japanese). Also, it clearly displays which Integration’s tool and which Skill was used, making it easy to verify the source.

Credit Usage

The footer of each AI response shows the credits used to generate that response.

OAuth Token Renewal

If an MCP tool’s authentication expires during conversation, you can immediately renew authentication by clicking the Re-authenticate button within the chat window without moving screens. After authentication is complete, you can continue the interrupted task immediately.

Install or Activate Required MCPs

In general chat, if an MCP tool required to handle the request is not installed yet or is disabled, a guide card is displayed in the chat window. You can install or activate the required Integration in the conversation flow and then continue with the same request without moving to a separate menu.

• If installation is required: click Install on the guide card and complete authentication or setup on the displayed installation screen.
• If it is disabled: turn on the switch in the guide card to activate the Integration.
• After installation or activation is complete, click Ask again to run the same request again.

Some Integrations can be installed directly in the chat window. Integrations that require separate manual setup must be configured first in the Integration menu or administrator settings according to the guide.

Previous Conversation Search

In general chat, you can search conversations from the past. When you ask AI to find something discussed in a previous conversation, a decision that was made, or a remembered keyword or identifier, it searches existing conversation titles and some related messages and presents the results.

Examples:

• “Find the conversation where we talked about the onboarding policy last time”
• “Find what conclusion we reached earlier about QPD-1234”
• “Continue searching from the previous period”

Search behavior

• When you enter a search term, the search starts with the most recent 30 days.
• If you include a date or date range, the search uses that range, but one search window can cover up to 30 days.
• If the results are insufficient or older conversations are needed, you can continue searching earlier periods in 30-day increments.
• The maximum searchable range is the most recent 93 days based on the retained search index.
• Search results include the conversation title, some related content, and a link to open the conversation.

Note: The current conversation is excluded from search results. Search result content comes from messages you previously entered, so open the original conversation to verify important decisions or numbers.

Tool Chaining (File Handoff Between Tools)

Files can be automatically passed between MCP, Skill, Sandbox, and My Drive to perform complex tasks. The AI uses internal temporary file addresses to automatically handle file transfers between tools, so users can perform multi-tool workflows with just natural language commands without any separate file management.

Supported chaining scenarios:

• MCP → Skill: Analyze/process files retrieved via MCP with a Skill
• Skill → MCP: Send files generated by a Skill to external services via MCP
• MCP → Sandbox: Execute/process MCP files in the Sandbox environment
• My Drive → MCP: Use files saved in My Drive with MCP tools

Usage examples:

• Retrieve a document from Google Drive, analyze it, create a PDF report, and send it via Gmail
• Process data collected via MCP with a Skill and save it to My Drive
• Process files from external services in the Sandbox and deliver the results back to the external service

Chat Sharing

You can share chat content with other users.

Click the Share button in the top right of the chat screen to open the sharing dialog.

Sharing scope settings:

• Private: Only you can access. (Default)
• Organization: Only members of the same organization can view.
• Public: Anyone with the link can view. This option appears only when an administrator enables Allow Public Conversation Links.

After selecting the sharing scope, click the Create share link button to generate a sharing link. The generated link can be copied with the Copy link button, or shared as a QR code via the QR code icon. Even if you change the sharing scope, the existing link is maintained, so there’s no need to share it again separately.

When an administrator turns off Allow Public Conversation Links, only Private and Organization remain available in the sharing dialog for both regular chat and agent conversations. If a conversation was already shared as Public, the dialog shows that its previous scope was Public and displays “Public links have been paused by an administrator setting.” You can still change the scope to Organization or switch to Private.

Note: The sharing link only includes conversation content up to the point of sharing. If new messages are added after sharing, a refresh button will appear, which you can click to update the link with the latest conversation content.

Caution: Please be careful not to share conversations that contain personal information.