pi-diesel-km

pi-diesel-km

Pi extension that displays model-aware LLM energy as an equivalent 5L/100km diesel-car distance.

It uses Pi session usage metadata plus a TypeScript port of EcoLogits request-energy math. When EcoLogits model metadata is available, estimates use model architecture and deployment data; otherwise the extension falls back explicitly.

Installation

pi install https://github.com/cgint/pi-diesel-km

Or during local development:

pi install /Users/cgint/dev-external/pi-diesel-km

Usage

/diesel-km       # Show detailed session energy and diesel-distance breakdown

The extension also updates a compact footer status entry such as 🚗~12cm after turns with assistant token usage.

Configuration

ECOLOGITS_MODELS_JSON=~/dev-external/ecologits/ecologits/data/models.json

If the metadata file is missing or cannot be parsed, the extension keeps working and marks unknown models as fallback estimates.

What it estimates

• Generated-token energy uses the EcoLogits-derived request-energy formula.
• Provider PUE values are mirrored from EcoLogits provider config.
• Diesel distance is an energy-content comparison: 5L/100km × 9.8 kWh/L = 0.49 kWh/km.
• Context-inclusive output is experimental and uses an Epoch-derived input-token heuristic.

This is not lifecycle CO₂ accounting and not tailpipe-emissions equivalence.

Development

npm install
npm run precommit

The package structure mirrors the other local Pi extension packages:

index.ts
src/
test/
README.md
LICENSE
AGENTS.md
NORTH-STAR.md
package.json
tsconfig.json
vitest.config.ts