Docs

Agent alerts

Long agent runs shouldn’t need babysitting. Agent alerts watch your AI CLI session and fire the moment the agent finishes a turn or stops to ask for input — so you can go do something else.

What fires

Pick any combination in Settings (⌘,) → Alerts:

Supported agents

Terminal Candy watches nine agents. Enable each with a single toggle in Settings → Alerts — no manual config editing.

AgentHow it’s watched
Claude CodeHook
Codex CLILog poll
Copilot CLIHook
Gemini CLILog poll
Cursor AgentHook
CodeBuddyHook
Kiro CLIHook
Kimi CLIHook
OpenCodePlugin

Hook agents get a tiny turn-complete hook written into their own config (e.g. Claude Code’s ~/.claude/settings.json Stop/Notification events). Toggling off removes it cleanly. If another tool already owns the hook slot, Terminal Candy wraps it transparently rather than clobbering it.

Log poll agents (Codex, Gemini) don’t expose hooks, so Terminal Candy tails their session logs (about every 1.5 s) and detects the turn-complete record — e.g. Codex’s task_complete event.

Plugin (OpenCode) registers a small plugin in opencode.json.

In every case the agent itself runs completely unmodified — the hook/poll only observes the “I’m done” signal.

Setup

  1. Settings (⌘,) → Alerts
  2. Toggle on the agents you use. A toggle only sticks if the agent is actually installed (Terminal Candy checks for its config directory).
  3. Choose your alert style — bounce, sound, bring-to-front.
  4. Run your agent in any Terminal Candy session and step away.

Alerts fire based on the agent’s own signals, so they work no matter which Terminal Candy tab or window the agent runs in.

Notes & limits