Conversation View

Read a running AI agent as a chat instead of a terminal: who is answering, where a long answer started, and which tool steps passed or failed.

An agent's terminal is built for the agent, not for you. A long session turns into a scroll of redrawn frames, tool output and diffs, and the answer you care about is somewhere in the middle of it.

The Conversation view shows the same session as a chat: your prompts, the agent's answers rendered as markdown, and its tool steps folded out of the way. The agent keeps running in its terminal the whole time; the view only changes how you read it.

Switching views

AI terminals carry a Conversation | Terminal toggle in their header. Click Conversation to read the session as a chat, and Terminal to go back to the live TUI.

The shortcut is Cmd+Shift+M on macOS and Ctrl+Alt+M on Windows and Linux (Toggle Conversation / Terminal). It can be rebound in keyboard shortcuts.

If the transcript can't be read, the Conversation tab turns amber and its tooltip says why, with a retry.

The session strip

The top of a conversation names the session you are reading:

the agent and the model it is running
the short session id: click it to copy the full id, for a resume command or a bug report
the message count
a Terminal button that opens the live terminal behind the conversation

With several agents open in one project, the strip is how you tell them apart before you reply.

Reading long answers

Answers sit in a narrow column, so lines stay readable on a wide monitor.

When the start of a long answer has scrolled off the top, a reading bar appears. It names the agent and the time of the answer, and Top of answer takes you back to its first line.

At the bottom, the jump button reads Latest when you have scrolled up, or shows how many new messages arrived while you were reading further back.

Tool steps

An answer's tool steps (file reads, searches, edits, shell commands) fold into one dot per step: green when the step succeeded, red when it failed. A twenty-step run fits on one row, and a red dot shows where it went wrong.

Expand the row for a compact list of the steps, and open any one with Details to see what it did.