Repository navigation
Simplest Possible Start
Three commands. No configuration, no GitHub login, no project directory. If you have used Claude Code on this machine, this shows what it cost.
brew tap subversivesoftwareorg/tap
brew install conscience
conscience report energy --allEvery Claude Code and Codex session on the machine from the last 30 days, grouped by model, with an energy estimate, its uncertainty, CO2 and water, and everyday comparisons chosen for the size of the number:
Conscience — Energy Estimate (30 day window)
╭──────────────────┬──────────┬──────────┬───────────────┬──────────┬─────────────╮
│ Model ┆ Tier ┆ Sessions ┆ Output Tokens ┆ Est. Wh ┆ Uncertainty │
╞══════════════════╪══════════╪══════════╪═══════════════╪══════════╪═════════════╡
│ claude-fable-5-1 ┆ large ┆ 8 ┆ 13.8M ┆ 86063.1 ┆ ±50% │
│ claude-opus-4-6 ┆ frontier ┆ 20 ┆ 4.4M ┆ 106414.0 ┆ ±100% │
╰──────────────────┴──────────┴──────────┴───────────────┴──────────┴─────────────╯
Totals
Estimated energy: 282932.4 Wh (range: 88259–477606 Wh)
Estimated CO2: 118.832 kg (at 0.42 kgCO2/kWh)
Estimated water: 509.3 liters (at 1.8 L/kWh)
Per day over 30 days: 9431 Wh, 3.961 kg CO2, 17.0 L water
Equivalent to (illustrative)
Energy ≈ 9.6 days of a US household's electricity (3–16)
source: EIA 2022, 10,791 kWh per household per year
CO2 ≈ 297 miles of gasoline-car driving (93–501)
source: EPA average passenger vehicle, 404 g CO2 per mile
Water ≈ 7.8 showers (2.4–13)
source: EPA WaterSense, 8 minutes at 2.1 gallons per minute
Every figure is an estimate from published research, shown with its range and its source. No provider publishes official numbers. The caveats printed under the table say exactly what the estimate does and does not cover.
- It reads only what is already on your machine: Claude Code's logs under
~/.claude/projects/and Codex's under~/.codex/sessions/. Nothing is sent anywhere. - It needs no GitHub token and no
conscience.yaml. -
--allmeans every project, so it works from any directory.
If it says no sessions were found, you haven't used Claude Code or Codex on this machine yet, or their logs live somewhere else.
From inside a project you work on with AI tools:
cd ~/code/your-project
conscience examineThe current directory is the project and its origin remote is the repository, so still no flags. You get what was analyzed, what deserves attention, and one question worth discussing. Then, when you want the questions with your team:
conscience reflect --interactiveGetting Started walks through both with the optional pieces: GitHub login, conscience.yaml, and CI.
cargo install conscience # any platform with RustOr download a binary from the releases page.
The eight commands
-
setup— what's configured -
examine— analyze a project, PR, or all projects -
report— github · ai · energy · tokens · authorship · attention · history · automation -
reflect— retrospective questions -
retro— aggregate saved reflections -
push— send a snapshot to a dashboard -
prune— remove a launcher behind failing automation -
du— what conscience takes up on disk;--tidyold snapshots
Guides
- Simplest Possible Start
- Getting Started
- Running a Team Retrospective
- Evaluating a Pull Request
- Understanding Your Attention Patterns
- Measuring Energy Cost
- Setting Up CI-CD
- Tracking Trends with a Dashboard
Reference