SHIPPED · v1.64.0

Conversation — read the running agent as chat

New

  • Switch any AI pane between Terminal and Conversation — a control on the pane shows the same live agent as a chat transcript instead of the raw TUI, without starting a second session. Your prompt still goes through the same Agent Input bar the terminal uses. You get "You" bubbles, markdown replies, tool rows, and — where the agent records them — thinking blocks. Load earlier pages history; Jump to latest returns you to the live end if you scrolled up. If a transcript never appears, a banner offers Terminal instead of leaving you on a blank chat.

  • Works with the coding CLIs you already run — Claude, Codex, Gemini, Kimi Code, Antigravity, Qwen, OpenCode, Cline, Grok, Hermes, Cursor, Pi, Amp, and Qoder all read as chat.

  • Follows an in-terminal resume — picking a past Claude conversation with /resume (or --resume) updates Conversation to that session, instead of waiting forever on the empty one you just left.

  • A compact composer under the transcript — one line that grows as you type, with Routing / Hierarchy / Pipeline on a single row, so the chat stays the main surface. Turn Conversation off in Settings → Terminal if you only want the live TUI.

  • The prompt-first layout is now called Conversation — same arrangement as before (wide center, Preview / Changes / Code on the right); stored layouts keep working.

Improved

  • Orchestration pickers stay inside the app — Builder, Start, and repair no longer open native OS menus that cover the dashboard. Search models, skills, projects, and seat terminals; leave a model blank for the agent's default or type a custom id.

  • Grok sits with your other AI accounts — Settings → Accounts lists Grok alongside Claude, Codex, Gemini, Qwen, and OpenCode, with credit alerts on the same row.

  • The compact project rail scrolls — Terminal Right, Terminal Center, and Chat no longer clip later project icons behind the status bar.

Fixed

  • Fixed Qwen and Gemini seats in a team or pipeline never receiving the brief — those agents were being launched with Claude's permission flag, which they reject; they now start with their own auto-approve flags and the prompt lands.

  • Fixed Terminal Right snapping terminals back to chat — choosing single, grid, or tabs in that preset is kept across project switches and restarts.

  • Fixed “Drop folder to add project” overlay staying on after a drop — the guide dismisses once the folder is added or selected.

  • Fixed phone Browser showing a blank page for desktop localhost — localhost:3000 and the other local spellings now proxy to the desktop app instead of the phone itself.

  • Fixed Conversation composer caret sitting in the middle of the text — the insertion point stays on the character you are editing, matching Terminal Mode.

  • Fixed Antigravity tabs never naming themselves or keeping a session — titles and Conversation show what you typed, not <USER_REQUEST> wrappers, and a restart can resume the same conversation.

  • 1DevTool Conversation view showing a live Pi agent session as a chat transcript, with tabs for Claude, Kimi, Cline, Cursor and Qwen agents and a compact composer carrying Routing, Hierarchy and Pipeline toggles
    1DevTool Conversation view showing a live Pi agent session as a chat transcript, with tabs for Claude, Kimi, Cline, Cursor and Qwen agents and a compact composer carrying Routing, Hierarchy and Pipeline toggles
  • Conversation transcript of a Claude Code agent with a You prompt bubble, a markdown reply, and collapsed Bash tool rows each showing a green check and a Details link, the Conversation toggle highlighted at the top right
    Conversation transcript of a Claude Code agent with a You prompt bubble, a markdown reply, and collapsed Bash tool rows each showing a green check and a Details link, the Conversation toggle highlighted at the top right
  • The Conversation and Terminal toggle on an AI pane, switching a running agent session between a formatted chat transcript and the raw terminal output
    The Conversation and Terminal toggle on an AI pane, switching a running agent session between a formatted chat transcript and the raw terminal output