Try JFrog
Guides
Contact SupportTry JFrog
  1. JFrog Boost

Token Savings with JFrog Boost

Understand how JFrog Boost compresses noisy command output for agents.

JFrog Boost prevents verbose test logs from flooding the context window your coding agent uses. Boost preserves the signal, including failures, summaries, counts, and next clues. But repetitive log noise is compressed before it reaches the agent. The full raw output remains retrievable when you need it.

What Boost Keeps

Boost preserves output that helps an agent act on a result:

  • Test failures and assertion messages
  • Pass or fail summaries and counts
  • Build errors and lint findings
  • Hints such as file names and line numbers tied to a failure

Everything else—progress lines, duplicate status rows, and verbose tool chatter—is candidates for compression.

Where Boost Helps Most

The noisiest commands your agent runs shrink the most. Common examples include test runs, build and lint logs, diffs, and container or cluster logs.

# noisy test runs
pytest -q
go test ./...
npm test

# build and lint logs
npm run build
npx tsc --noEmit
ruff check .

# diffs, logs, and debug output
git diff
docker logs app
kubectl logs deploy/api

For tools without a built-in filter, add custom rules in TOML Filters for JFrog Boost.

Measuring Token Savings

Boost tracks raw versus filtered size, and it also estimates saved tokens per command. Run boost report for aggregate savings across your sessions.

boost report

You can configure reporting defaults as well as the calculations used for cost or emissions estimates. For more information, see Configuring JFrog Boost.

Emission Savings

Saving tokens directly reduces the computing power you consume. Boost translates raw token savings into an estimated emissions impact using a configurable calculation (default 0.21 kg CO₂e per 1 million saved tokens).

$ boost report

● Saved 1.7M tokens of terminal noise.
● You avoided 359 g CO2e. ≈ 0.21 kg CO2e / 1M saved tokens.
📘

Note

The estimate is a directional sustainability signal, not audited carbon accounting. It turns context saved into a concrete reminder that less AI exhaust reaches the atmosphere. Adjust the calculation with [report] co2e_kg_per_million_tokens in .boost/config.toml.

Related Topics