Custom agents, and terminals you can select

**Custom agents — plug any coding CLI into the workspace** — register a fork, a brand-new coding CLI, or a command you already launch by hand as a first-class agent, without waiting for a 1DevTool release. Open **Settings → Terminal → Commands**, mark the command as **AI agent**, and expand **Agent integration**.

**Give it a name, an accent colour and a logo** — the logo is downscaled to 64px and stored with your settings; without one, the first letter sits on the accent colour (the default is #7C6AE0).

**A custom agent shows up everywhere a built-in one does** — Add Terminal, the empty-state launcher cards, Change AI, Prompt History, Resume and Memory all render it with its own logo and colour.

**Sessions survive a restart** — point 1DevTool at where the CLI writes its transcripts (**Pi-style sessions**, **Claude-style projects**, or a flat JSONL folder), then press **Test** to see real sessions listed before you save. **Open** reveals the folder in your file manager.

**Resume, continue and the agent's own picker are templates you fill in once** — `{cmd} --session {id}` to resume by id, `{cmd} -c` to continue the latest, and either a slash command typed into the running TUI or a CLI flag such as `{cmd} -r` for the agent's own picker.

**Pick how the terminal treats the screen** — **Full-screen TUI** lets the agent own the alternate screen and its own scrolling (OpenTUI-style, the safe default); **Inline transcript** blocks the alternate screen so scrollback stays in 1DevTool, the way Claude and Codex behave.

**Name the memory files yourself** — list the per-project files (for example `AGENTS.md`) and an optional global file that the Memory panel should read and edit for that agent.

**Custom agents sit in the same Settings list as the built-ins** — same Show toggle, same drag order. Built-in names are reserved, so you cannot register over Claude, Codex or any CLI 1DevTool already ships. Deleting the command asks whether to drop the registration; terminals already open keep working.

**Select terminals, then merge or close them** — right-click a terminal and choose **Select**. Checkboxes appear on every visible tab and pane, including mosaic tiles, and the one you clicked starts checked. Checking a box never switches the active terminal.

**Merge several sessions into one agent** — the action bar combines the selected transcripts into a single prompt and hands it to an existing AI terminal or a new one. The dialog lists each source with its message count, and warns when a terminal *could not be attributed to a session* so you know exactly what will and will not contribute context. Optionally **close the merged terminals afterwards**.

**Close a whole selection in one step** — or right-click a project and choose **Close All Terminals**, with a confirm step that still leaves **Reopen Closed Terminal** intact. **Select all**, **Done** and Escape leave selection mode.

**Switch worktrees from the status bar** — when a project has linked worktrees, the status bar names the active checkout with its colour and primary/locked/dirty marks, and switches without opening Git Client. Switching brings forward a terminal that already belongs to that worktree instead of leaving you on a tab from another checkout; a worktree with no visible terminals shows its own empty state.

**Markdown and the terminal are real side-by-side panes** — Reading Mode no longer covers the document with a floating dock. Pick **5 : 5**, **3 : 7**, **7 : 3**, markdown above the terminal, or terminal above markdown; drag the sash to resize, and the layout you pick is remembered.

**A diagram lightbox you can actually use** — resize the source pane, collapse it to a rail, and double-click the sash to reset.

**Switching layouts no longer rebuilds every terminal** — grid, columns, rows, single, vertical tabs, chat and vibe keep the panes you already had open instead of flashing *Opening terminal* on each one.

**Jumping back to a recent terminal is instant** — a pane you just left is still there, with its scroll position and selection, instead of replaying from scratch.

**More terminals stay sharp at once** — GPU rendering covers a larger set of panes, and an off-screen pane gives its slot back so a busy workspace does not fall over to a blurry fallback.

**A workspace with dozens of terminals stays responsive** — idle CPU and memory stay flat instead of climbing as you keep adding agents.

**Mission Control no longer steals the composer** — when it auto-reveals beside Agent Input it borrows only the width the map needs, and gives the rest back the moment you switch panels or drag the splitter.

**Diagrams match the rest of the workspace** — Mermaid charts in the markdown preview and graph editor use a quieter palette, and loud author colours are toned down so nodes stay readable in both themes.