AgentRadar · on the App Store
Free · one-time Pro unlock, no subscription
macOS 13.0 or later
AgentRadar alternatives: how do people watch their agents?
If you run several AI coding agents at once, you already have a way to check on them. The question is what that way costs you — in setup, in screen space, and in trust.
Side by side
| AgentPet (open source) | Tail logs in terminal windows | Activity Monitor | AgentRadar | |
|---|---|---|---|---|
| How it observes | Hook injection into the agent | You read raw log files | Process-level resources | Zero injection — reads the session records agents already save |
| Form | Desktop pet | Terminal windows | System utility table | Menu-bar panel (+ optional desktop window) |
| Answers "who is waiting for me?" | Yes | You infer it from the log tail | No | Yes — first-class status with notifications |
| Token / cost tracking | No | No | No | Yes — by agent, project and model, with editable prices |
| Setup | Install hook + pet | Write your own scripts | Built in | Install, open, done — adapters included |
| Modifies agent files | Yes (by design) | No | No | Never — read-only by construction |
The honest trade-off
AgentPet's hook injection buys precision — the pet knows exactly when an agent is blocked. AgentRadar's zero-injection reading is less invasive: nothing is installed into the agent, nothing is modified, no account, no uploads. For most statuses (working, done, token usage) the read-only signal is exact; for "waiting for input" without a hook it is a heuristic. If you want that one signal to be exact, the optional hook closes the gap — everything else works without it.
When the alternatives are enough
- One agent, one window: if you only ever run a single agent and never lose track of it, the menu bar is already your panel.
- Terminal-native workflows: if you live in tmux and enjoy tailing logs, a panel adds little — though you still lose the cost view.
- Resource troubleshooting: Activity Monitor remains the right tool for CPU and memory questions; AgentRadar is about agent state and spend, not processes.