A coding agent running in a terminal is a strange thing to supervise. It is doing a lot, most of it fast, and the only window into it is a scrolling log in a terminal you have probably switched away from. You come back to find it finished, or stuck, or thirty files deep in something you did not ask for. The Agents droplet gives that work a place at the top of the screen, the way a timer or a download has one.
Where the information comes from
Every one of these agents writes a local session log as it works. Claude Code writes its transcript under your home folder. Codex CLI writes rollout logs. Cursor, Antigravity and OpenCode each keep their own. The Agents droplet watches those files, on your disk, and turns the entries into a live view. It does not talk to the agent, does not hook into the terminal, and does not contact any service. If you have no agent sessions, it shows nothing.
That design has two consequences worth stating plainly. It is fully local: nothing about your code or your prompts leaves the Mac because of this droplet. And it only ever sees what the agent already wrote down, so it cannot see more than the agent's own log does.
The notch pill and the shelf card
A small live indicator. The pill beside the notch says an agent is working and moves while it does. A ring shows progress through the current turn, and it starts moving the moment the HUD unfolds. Glance up and you know whether it is still going without switching to the terminal.
The expanded view. What the model is thinking, in its own words. The current tool call: reading a file, running a command, editing. The file being edited and the line count changed so far. Each tool call appears as it happens, so a long session reads as a timeline rather than a wall of text.
Read only, and less. With Droppy for iPhone paired, Agents shows which agents are running on the Mac and what each is doing, with no file paths and no prompts. Enough to know a run you left at the desk is still moving. It can be a Home Screen widget.
Multiple sessions at once each get their own entry. Switching between them is a tap.
Which agents, and how they differ
| Agent | Source | What the card shows |
|---|---|---|
| Claude Code | Local session transcript | Thinking, tool calls, file, lines changed |
| Codex CLI | Rollout logs | Turn tracking, tool calls, file, lines changed |
| Cursor | Local agent session | Tool calls, file, lines changed |
| Antigravity | Local agent session | Tool calls, file, lines changed |
| OpenCode | Local agent session | Tool calls, file, lines changed |
The depth varies with what each agent writes down. Claude Code's transcript includes its reasoning, so the card can show what the model is thinking. Others log the tool calls and edits and not much narration, so the card shows the calls and edits. The droplet never invents a status; if the log does not say, the card does not either.
Why this belongs in the notch
Droppy's Live Activities are for things that are happening now and will end: a timer, a recording, a download, a meeting. An agent run is exactly that shape. It started, it is progressing, and it will finish or need you. Putting it in the same place as the other live things means you form one habit for all of them: glance up. It also means the run does not need its own window open, which is the difference between supervising an agent and being held hostage by a terminal tab.
Running several at once
People who use agents seriously often run more than one: Claude Code on a refactor in one worktree, Codex on tests in another. The card lists each session and lets you switch. The pill reflects whichever is most recently active. That is the moment the phone view earns its keep: from the sofa, the Agents droplet on the iPhone says both are still working, and you can go back to whatever you were doing.
Setting it up
Switch Agents on in the Droplet Store. There is nothing else to configure: start a session in any supported agent and the pill appears. If you use an agent that is not on the list, the droplet ignores it rather than guessing, and the Droplet Store page is the place to ask for it. The Agents droplet shipped in Droppy 14.2 beta 4 and has picked up agents since.