Forks from opencode-throughput

Why fork? I want the simpler realtime stats bar, instead of complete.

opencode-throughput

Real-time LLM performance monitoring plugin for OpenCode. Tracks TTFT, TPS, latency, token usage, and cost with a live readout in the prompt status bar. Supports both OpenCode 1.x (V1 TUI plugins) and OpenCode 2.x (V2 CLI plugins) from one package.

Features

• TTFT (Time To First Token) — measures how fast the model starts generating
• TPS (Tokens Per Second) — generation throughput during streaming
• Total Latency — end-to-end request time
• Token Usage — input, output, reasoning, cache read/write
• Cost Tracking — per-request and cumulative cost
• Status Bar Readout — live TTFT · ↓tokens · TPS while streaming, plus a per-turn summary when idle

Installation

OpenCode 2.x — add to ~/.config/opencode/cli.json:

{
  "$schema": "https://opencode.ai/v2/cli.json",
  "plugins": ["@jumpmaster/opencode-throughput"]
}

OpenCode 1.x — add to tui.json:

{
  "$schema": "https://opencode.ai/tui.json",
  "plugin": ["@jumpmaster/opencode-throughput"]
}

OpenCode installs the plugin from npm at startup. Requires opencode >= 1.18.0; OpenCode 2.x needs no extra configuration.

Usage

Once installed, the plugin runs automatically. While a model generates, the prompt status bar shows the live readout:

TTFT 1.2s · ↓1.2k tok · 58.3 tok/s

When idle it shows the last turn's summary:

TTFT 2.1s · ↓1.5k tok · 57.3 tok/s

• TPS is generation throughput: it excludes prefill (TTFT) and tool-execution time
• The turn summary resets on each new user message / execution
• While the next prompt is still in prefill (no first token yet), the previous turn's token and TPS values stay on screen in muted color, so nothing blanks out

How It Works

The plugin is a TUI plugin (@opentui/solid + solid-js) with two implementations behind one entrypoint:

• V1 (src/v1.tsx) subscribes to message.part.updated, message.part.delta, session.next.*.delta, and message.updated, and renders into session_prompt_right / home_prompt_right
• V2 (src/v2.tsx) subscribes to session.step.started, session.text.delta, session.reasoning.delta, session.tool.input.delta, session.step.ended, and session.execution.*, and renders into prompt.footer.status / home.footer.status

Both share the pure TTFT/TPS helpers in src/stats.ts. V1 reads the first token time and the completed message totals; V2 uses the step start time as the TTFT anchor and session.step.ended for real token usage per step.

Development

Local development

V1 — add to tui.json (TUI plugins are loaded from tui.json, not opencode.json):

{
  "$schema": "https://opencode.ai/tui.json",
  "plugin": ["file:///absolute/path/to/opencode-throughput/src/index.ts"]
}

V2 — add to ~/.config/opencode/cli.json:

{
  "$schema": "https://opencode.ai/v2/cli.json",
  "plugins": ["file:///absolute/path/to/opencode-throughput/src/tui.tsx"]
}

Edits to src/** only take effect after restarting opencode. Keep exactly one registration (published package or local file) while developing.

Build

bun install
bun run build
bun run typecheck
bun test

bun test includes a render test that drives the V2 slot with synthetic events via @opentui/solid's testRender.

Publish

npm login          # once
bun run release    # bump + build + typecheck + test + npm publish

bun run release [patch|minor|major|x.y.z] [--dry-run] [--yes] [--tag=beta]

License

MIT