Draw your agent team

**Charts is now one library for every org chart and pipeline** — it lives under **Set up** in the orchestration dashboard and does exactly one thing: draw the relationships between your terminals, who reports to whom and who hands work to whom. A **New chart** dialog picks the shape (Hierarchy, Pipeline, or the Router/Defaults table) and how you start it — **Use AI** or **Do it yourself** — instead of hiding the choice behind a mode switch.

**Start from the terminals you already have open** — with two or more AI terminals running, Agent Input offers **“Draw how these N terminals work together”**, and Charts' **Chain my open terminals** card drops exactly those terminals into a pipeline, already assigned.

**Describe your team in plain language** — a ✨ **Describe with AI** popup with a real text box, clickable example prompts and a three-step guide (*Describe it → We draw a draft → You press ▶ Start team*) drafts the chart for you. A **Generate with** chip picks which installed AI CLI does the drafting, and shares one preference with the Draw canvas. Drafting only replaces the canvas — ⌘Z brings the old chart back, and it never starts a team.

**Start screens that show your options instead of hiding them** — a blank chart, ready-made org charts and pipelines, your saved charts, and an explicit **Convert my pipeline / Convert my org chart** card, so converting a draft is a choice you make rather than a gate you pass through. Opening a saved chart over unsaved edits now previews what is on the canvas first, with **Save edits first**, **Cancel** or **Discard and load**.

**One Start per screen** — while the seat table is open, the confirm at its foot is the only Start button, so you're never choosing between two controls for the same act.

**The Router is now Defaults**, moved to **Advanced** — it is consulted at exactly two moments: work you never named an agent for, and a run that failed. Naming an agent always wins.

**Fallback chains** — give any type of work an ordered list of backups, choose which failures trigger them (launch error, provider unavailable, rate limit, quota exhausted, no response), and decide whether the chain advances on its own or waits for you. Active teams show the hop that actually happened — *Codex rate-limited → Claude*.

**Defaults fills itself in on first run** — a starting table is seeded from the AI CLIs you really have installed, instead of an empty grid. After enough runs it can also point out a type of work that keeps failing on one engine and propose a change; it never edits your table by itself.

**A routing board you can see whole** — work types on the left, the agent that takes them on the right, sized to the space it has.

**Resource Manager — see what's using your machine** — click the memory reading in the status bar for a per-process breakdown: 1DevTool's own processes (Main, Renderer, GPU) and every terminal, named the way its tab is. Live CPU sparklines run on each row while the panel is open, system memory, swap and CPU load sit underneath, and **Clear terminal buffers** releases cached scrollback so you can watch the number come down.

**Cursor and Grok now read as chat** — a Cursor terminal shows its parsed transcript in Conversation and Reader Mode instead of a raw screen scrape, its sessions appear in Resume AI Sessions, and it resumes after a restart like every other agent.

**Multiple-choice questions with mockups keep their shape** — when an agent's question carries side-by-side ASCII previews, the transcript draws each option's mockup in its own monospace box with the columns intact, instead of a wrapped blob of JSON.

**Ctrl+click a file path an agent printed and get the right file** — paths now resolve against the terminal's own working directory (a git worktree included) before the project root, and only open when the file is really there.

**Terminal mentions look like the terminals they name** — an @terminal chip carries that agent's logo, and clicking it focuses that terminal.

**Drag a search result straight into a prompt** — file-search results in the explorer are draggable now, so a match can be dropped onto Agent Input as an @mention just like a file from the tree.

**A real New terminal button** — the tab strip's bare “+” is now a labelled control showing your actual shortcut, including one you remapped yourself.

**Cursor and Cline own their quota pill** — a focused Cursor or Cline terminal shows its own state and a link to that provider's usage page, instead of borrowing another provider's percentage.

**Projects are recognisable everywhere** — the project switcher draws each project's avatar, colour and emoji the way the sidebar does, and the same chip now marks Tasks cards, the agent roster, and assignment targets, so a task from another project reads at a glance.

**Route & run shows its work first** — batch assignment, single **Route & run**, and Agent Input's Orchestrate all open the same preview of which agent each task will go to before anything starts.