Chat view

Every agent pane can switch between its terminal and a chat view. The chat view is a layer over the same running CLI: nothing is started a second time, and no extra request goes to your provider. PaneScale reads the session file the CLI writes anyway and types your input into the terminal, exactly as you would.

#Switch views

Switching is instant. The session and the view stay in memory as long as the pane lives, so you can flip back and forth at any time — both views always show the same session.

The chat view reads the full history of Claude Code, Codex and Kimi from their own session files. Other agents keep the terminal as their main view.

#What you see

• Your messages and the answers, rendered with Markdown and highlighted code blocks. Copy message copies an answer.
• The work log — every file read, edit, command and search, grouped and expandable, failed steps marked. Thinking appears as a short Thinking line while it happens.
• Plans as a card — Proposed plan, with Implement and Revise. Task lists show their progress: Tasks: 2 of 5 done.
• Subagents of a Claude Code session, each with its prompt and output. Open one to follow it; Back to main chat or Esc returns.
• Background work the agent started, marked Running in the background.

Long conversations load from the end; Load older messages fetches earlier parts, Jump to end follows the live answer again.

#Write and send

Menus such as Claude Code’s /config or /mcp appear inside the chat view and are operated with keys and the mouse, just like in the terminal.

#While the agent works

A message you send while the agent is busy goes into a queue and is sent after the current step. Send now (Ctrl+Enter) uses the CLI’s own way to get it in right away: Claude Code interrupts the current step, Codex adds it to the current turn. The tooltip of the button says which.

#Questions and approvals

When the CLI asks something — a permission for a command or an edit, a plan to approve, a question with options, Trust this folder? — the chat view shows it as a card with buttons: Allow, Reject and the CLI’s other options, or the options of the question with room for your own answer. Your choice goes to the CLI as the same keys you would press in the terminal. Until it is answered, new messages wait: Answer the approval first to continue.

#Model, effort and permissions

The bar under the input shows the session’s model, effort and permission mode. Changes made here apply to this session only — your default for new sessions stays as it is. Set as default … opens the CLI’s own menu if you want to change the default too. If the agent is busy, an effort change waits until it is idle (Switch to “High” is pending). Agents without an effort setting say so.

#Context ring

The small ring in the bar shows how much of the context window is used. Hover it for the exact share, the tokens and — where the CLI reports it — the cost at API prices. When the CLI compacts the conversation, the chat view shows Compacting conversation.

#Good to know

• The chat view never runs a second agent. Everything you do in it happens in the terminal session underneath, and the status of the pane stays the same in both views.
• It only reads local files. Nothing about your conversation leaves your computer through PaneScale.
• If the session file can’t be read for a moment, the view retries by itself.
• For chats that don’t need a terminal at all, use Chat mode.