Hud

A plugin for OpenCode that displays token streaming metrics at the end of each conversation.

[!Warning] This plugin is still in developping. Maybe not available. 🐛

Metrics Displayed

Metric	Description
⚡ Avg TPS	Average tokens per second (includes reasoning tokens)
TTFT	Time To First Token — latency from request to first response token
Total tokens	Cumulative token count from API (output + reasoning)
Elapsed time	Wall-clock time since the first token

Example toast:

⚡ 42.5 t/s  TTFT 312ms  [639 tok / 15.0s]

Installation

From npm

Add the plugin to your .opencode/opencode.json:

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

Local Development

This plugin auto-loads when you run opencode from this directory, because it lives in .opencode/plugins/.

Configuration

Create a config file at ~/.config/opencode/opencode-hud.json:

{
  "enableLogging": true,
  "logFilePath": ".opencode/hud-debug.log"
}

Option	Type	Default	Description
enableLogging	boolean	false	Enable debug logging
logFilePath	string	.opencode/hud-debug.log	Path to log file

Development

# Install dependencies (Bun required)
bun install

# Run tests
bun test

# Type-check
bun run typecheck

# Build
bun run build

Project Structure

opencode-hud/
├── .opencode/
│   ├── opencode.json          # OpenCode project config
│   └── plugins/
│       └── hud.ts             # Plugin entry (auto-loaded by OpenCode)
├── src/
│   ├── index.ts               # Plugin main entry — event handlers
│   ├── config.ts              # Configuration management
│   ├── logger.ts              # Configurable logging
│   ├── types.ts               # TypeScript interfaces
│   ├── metrics.ts             # Token estimation, duration formatting
│   └── display.ts             # Toast formatting and emission
├── tests/
│   ├── metrics.test.ts        # Unit tests
│   └── integration.test.ts    # Event flow tests
├── dist/                      # Build output
├── package.json
└── tsconfig.json

Architecture

The plugin listens to OpenCode events and displays metrics when a conversation ends (session.idle).

Event Flow:

1. message.updated (user) → Record request start time
2. message.updated (assistant) → Store message with tokens info
3. message.part.updated → Track first token time, count tokens
4. session.idle → Calculate metrics and show toast

Token Counting:

• Primary: Uses API-provided tokens.output + tokens.reasoning
• Fallback: Estimates from text length (length / 3)

Key Design Decisions:

• requestStartTime set on user message to capture full TTFT
• streamingStartTime set on first assistant token (lazy init)
• Token count includes reasoning tokens for accurate TPS
• Uses performance.now() for high-precision timing