$ token-savings
Token Savings
The anatomy of abloated context window
Repetitive terminal noise balloons the loop. Here is how Boost trims it.
This is your session with Boost on. Toggle it to see the raw, context-flooding output it just hid.
What Boost keeps
Boost preserves the signal: failures, summaries, counts, and next clues. The repetitive stream gets compressed before it spends context. If a filter hid too much, run boost retrieve <id> — Boost monitors those retrieve events and can disable the filter locally. Prefer retrieve over bypassing Boost.
Where it helps most
The noisiest commands your agent runs (tests, builds, and logs) shrink the most.
# noisy test runs
pytest -q
go test ./...
npm test
# build & lint logs
npm run build
npx tsc --noEmit
ruff check .
# diffs, logs & debug output
git diff
docker logs app
kubectl logs deploy/apiMeasure what you saved
Boost tracks raw versus filtered size and estimates saved tokens per command. Run boost report for aggregate savings, or add a TOML filter for custom tools.
# sustainability
Emission Savings
Every saved token is energy you didn't burn. Boost translates raw token savings into an estimated emissions impact, so the work your agents skip becomes a number you can share with your team.
● Saved 1.7M tokens of terminal noise.
● You avoided 359 g CO2e. ≈ 0.21 kg CO2e / 1M saved tokens.
Less noise → less compute → less AI exhaust in the atmosphere.
The estimate uses a configurable kg-CO2e-per-million-saved-tokens heuristic. It's a directional sustainability signal, not audited carbon accounting, but it turns context saved into a concrete reminder that less AI exhaust reaches the atmosphere.