OpenCode Savings Tracker

Track your local inference savings vs API costs in real-time. See how much you're saving by running models locally instead of paying API rates.

97% cheaper at home! Local inference costs ~$0.01 vs $0.30/M tokens on the API.

Features

• Automatic tracking - Plugin tracks every request silently
• Cache-aware - Tracks prefix caching (fresh vs cached tokens)
• Simple commands - /savings to see your savings
• Easy install - Copy files, restart, done

Installation

1. Install the plugin

mkdir -p ~/.config/opencode/plugins
cp src/index.ts ~/.config/opencode/plugins/savings-tracker.ts

2. Install the commands

mkdir -p ~/.config/opencode/commands
cp .opencode/commands/*.md ~/.config/opencode/commands/

3. Restart opencode

# Exit and reopen opencode, or restart the daemon

4. Test it

opencode run "hello"          # Generate some tokens
opencode run "/savings"      # See your savings!

Commands

Command	Description
/savings	Show savings summary
/savings-reset confirm: true	Reset all tracking data

What It Tracks

Token Type	Rate	Notes
Fresh input	$0.30/M	New tokens processed
Cache read	$0.06/M	From prefix caching
Output	$1.20/M	Generated tokens

Your local cost: Just electricity (~$0.0001/M tokens at $0.12/kWh)

Example Output

Savings Tracker Summary
======================
Period: 0.5 days (since 2024-04-24)

Usage:
  Total requests: 25
  Total input tokens: 3,456,789
    - Fresh: 2,123,456
    - Cache read: 1,333,333
  Total output tokens: 45,678

Costs:
  minimax/nvfp4 API: $12.34
    - Cache read: $0.08
    - Fresh input: $0.64
    - Output: $0.55
  Local inference: $0.01
  -------------------------
  Net savings: $12.33 (99% cheaper at home)

Configuration

Create ~/.local/share/opencode-savings/config.json to customize:

{
  "providers": ["llama-*", "minimax-*"],
  "baseline": {
    "provider": "minimax",
    "model": "nvfp4",
    "inputCostPer1M": 0.30,
    "outputCostPer1M": 1.20,
    "cacheReadCostPer1M": 0.06
  },
  "gpus": [
    {
      "wattage": 275,
      "promptTokensPerSecond": 1000,
      "outputTokensPerSecond": 43,
      "costPerKwh": 0.12
    }
  ]
}

Config Options

Option	Default	Description
providers	["llama-*", "minimax-*"]	Provider patterns to track
inputCostPer1M	$0.30	Input token rate
outputCostPer1M	$1.20	Output token rate
cacheReadCostPer1M	$0.06	Cache read rate
wattage	275	GPU wattage
promptTokensPerSecond	1000	Input speed
outputTokensPerSecond	43	Generation speed
costPerKwh	$0.12	Electricity rate

Requirements

• OpenCode with @opencode-ai/plugin and @opencode-ai/sdk
• vLLM with prefix caching enabled:

  --enable-prefix-caching
  --enable-prompt-tokens-details
  

• Or llama.cpp server (KV cache enabled by default)

Troubleshooting

No cache tokens showing?

• Make sure vLLM has --enable-prompt-tokens-details flag
• Restart the vLLM server after adding flags

Plugin not loading?

• Check for syntax errors: opencode run "hello" 2>&1 | head -20
• Verify plugin path exists: ls ~/.config/opencode/plugins/

Reset tracking data:

opencode run "/savings-reset confirm: true"

Future Enhancements

See GitHub Issues for:

• Track savings over time with charts
• Show cache hit rate percentage