Chapter 4: Chat Basics — every detail of the input box, message area and branches

This chapter guides you through the features of the chat page and what the information shown in each area means.

Reading time: about 10 minutes.

---

4.1 Page layout

The chat page is divided into three areas:

• Top: conversation title, current model, workspace directory, workspace directory picker, and the new-conversation button.
• Middle: the message display area.
• Bottom: the input box.

4.2 The bottom input box

From left to right, the buttons in the input box are:

1. No-approval mode button
2. Reasoning depth adjustment button
3. Fullscreen input button
4. Context window details button
5. Bottom action panel popup button

The send button is only shown after you type a message or attach a file — it stays hidden otherwise.

4.3 Conversation title

The conversation title reflects the topic of the first message in this conversation, along with its configuration. You can tap the title area to expand the full conversation metadata, including:

• Conversation title
• The model currently used by this conversation
• Workspace directory
• Conversation ID

4.4 Workspace

When editing and processing code files, it's recommended to create and edit them inside the workspace directory. Inside the workspace, better tools are available, and no file-management permission is needed (if you prefer not to grant the app full file access).

4.5 Creating a new conversation

Tap the top button to start a new conversation. Old conversations are kept in history and won't be lost until you delete them manually. Conversation management (archiving, deleting, projects) is covered in Chapter 5 "Managing Conversations".

4.6 Message display area

The message display area shows the complete:

• User messages
• AI chain of thought
• Tool call details
• AI messages

Common actions:

• The download button at the top-right of tables: export the table.
• tok/s at the bottom-left of AI messages: token output speed.
• The bottom-right of AI messages: the menu button.

4.7 Turn jumping

After a multi-turn conversation, you can swipe from right to left on the chat page to pull out jump buttons, which let you quickly jump to:

• The previous or next turn of the conversation
• Straight to the top / bottom

4.8 Message actions

• Long-press a user message: choose to copy the message.
• Not satisfied with the generated content: choose to re-edit and regenerate.

4.9 Tool call details

After the AI calls a tool, the tool call details appear in the chain of thought. You can tap to view the details and see the complete tool call JSON data. For the tool system itself, see Chapter 11 "Agent Tools".

4.10 Conversation branches

If you want to explore messages generated by different prompts without destroying the old ones, you can create a branch here. A branch keeps everything before the branch point, and the old conversation is not deleted.

4.11 Reasoning depth

If your model supports adjusting reasoning depth (also called thinking depth), you can set the level here.

4.12 Web search

The AI uses its built-in webview to search the internet after judging whether it's needed. You can manage the blocklist of sites the AI may visit in Settings → Web search.

External search services are not supported yet. The full web-search pipeline is in Chapter 11, 11.7.

4.13 Instruction controls

Typing a special symbol at the start of the input box steers the AI to read certain content:

• Type #: steer the AI to read a specified file.
• Type /: steer the AI to read a specified Skill (Skills are covered in Chapter 18).

---

End of chapter · Next: Managing Conversations — history, archive and projects