Token Monsters

Token Monsters adds a token usage sidebar to OpenCode. It shows the current context size, cache hit percentage, session totals, current prompt totals, prompt/tool/file breakdowns, and tool-definition overhead.

All numbers are approximate and counted with gpt-tokenizer using the OpenAI o200k tokenizer. For Claude, Copilot, and other non-OpenAI models, treat the numbers as a ballpark.

Install

Add the plugin to both OpenCode config files.

~/.config/opencode/opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": [
    "@aundal/opencode-token-monsters@latest"
  ]
}

Restart OpenCode after saving. OpenCode installs npm plugins automatically at startup.

~/.config/opencode/tui.json

{
  "$schema": "https://opencode.ai/tui.json",
  "plugin": [
    "@aundal/opencode-token-monsters@latest"
  ]
}

Install with OpenCode

Paste this prompt into OpenCode:

Install the OpenCode plugin @aundal/opencode-token-monsters@latest into my global opencode.json and tui.json plugin arrays. Preserve my existing config, validate config shapes against https://opencode.ai/config.json and https://opencode.ai/tui.json, and tell me to restart OpenCode when done.

Usage

Token Monsters appears in the session sidebar.

Use /tokenmonster to fold or unfold the sidebar panel.

You can also click the Token Monsters header to fold or unfold it with the mouse.

Views

• Session: Total shows everything captured since the session started.
• Session: Aktuel shows the last request only.
• View: Prompts groups usage per prompt turn.
• View: Tools aggregates tool and file usage.

Package

The package exports one OpenCode plugin module with both server and TUI hooks. The server hook writes token capture data, and the TUI hook renders the sidebar.

End users only install the npm package in opencode.json and tui.json. No build step is needed for end users. The published package ships the compiled files.

Development

bun install
bun run build
npm pack

Publish

bun run build
npm pack
npm publish --access public

License

MIT