dumb-zone-alert

Get notified before your LLM loses its mind.

An OpenCode plugin that tracks token consumption per session and fires desktop alerts when you approach the dumb zone: the point where a context window is so saturated that model quality degrades.

---

What is the Dumb Zone?

LLMs have a fixed context window. As the conversation grows, the model has to juggle more and more tokens, and past a certain point it starts dropping the ball: forgetting earlier instructions, hallucinating, repeating itself. That degraded region is the dumb zone.

The problem is that there's no built-in warning. You only notice when the output quality tanks. This plugin watches your token consumption in real time and alerts you at configurable thresholds so you can decide to start a fresh session before things go sideways.

---

Features

• Three alert tiers — 50 %, 75 %, and 100 % of your configured threshold
• Desktop notifications via notify-send with urgency levels (low, normal, critical)
• TUI toasts displayed inside OpenCode's terminal UI (with graceful fallback if unavailable)
• Per-session deduplication — each threshold fires at most once per session, no spam
• Per-project context — the notification title includes the current project name
• Fully configurable — adjust the token threshold, ratios, urgency levels, and messages

---

Prerequisites

• OpenCode installed and working
• notify-send available on your system (Linux with libnotify)
  • Install on Debian/Ubuntu: sudo apt install libnotify-bin
  • Install on Arch: sudo pacman -S libnotify
• macOS / Windows: notify-send is not natively available. Desktop notifications will silently fail; TUI toasts will still work.

---

Installation

1. Copy or clone dumb-zone-alert.js on your machine, e.g.:

   ~/.config/opencode/plugins/dumb-zone-alert.js
   

2. Restart OpenCode. The plugin loads automatically on startup.

---

Configuration

All configurable values are at the top of dumb-zone-alert.js.

TOKEN_THRESHOLD

The total token count that represents your model's effective limit. Tiers are calculated as percentages of this value.

const TOKEN_THRESHOLD = 100_000; // adjust to your model's context window

TIERS

Defines the ratio, visual level, notify-send urgency, and display label for each alert tier.

const TIERS = [
  { ratio: 0.5,  level: "info",    urgency: "low",      label: "50 %"  },
  { ratio: 0.75, level: "warning", urgency: "normal",   label: "75 %"  },
  { ratio: 1.0,  level: "error",   urgency: "critical", label: "100 %" },
];

MESSAGES

The body text of each notification, keyed by ratio.

const MESSAGES = {
  0.5:  "We're half way to the dumb zone!",
  0.75: "Be careful, the dumb zone is about to be reached!",
  1.0:  "You reached the dumb zone, you should start a new session.",
};

---

License

MIT — see LICENSE file.