Get Started
Boost overview
Boost is a single binary that sits between your coding agent and the terminal. It compresses noisy command output, traces important workflows, and surfaces performance signals — so agents spend tokens on signal, not noise.
What you can do
Here are the main capabilities Boost provides out of the box.
Reduce token consumption across 100+ commands
Every command agents run — git diff, cargo test,
docker build — returns raw terminal output the model has to read.
Boost filters that output at the source, delivering 60–90% fewer tokens with more accurate, signal-rich context.
boost git diff # compact per-file diff
boost vitest run # failures only, 99% reduction
boost docker build # step summary + cache stats Deep traces of builds and tests
Boost records an OpenTelemetry span for every command, including timing, exit codes, and token savings. For Docker builds it traces every BuildKit vertex; for test runners it extracts pass/fail/skip counts. This gives you visibility into performance and stability across local and CI workflows.
boost cargo test # per-test timing + flaky detection
boost next build # route metrics + build duration
boost docker build # per-layer cache hit/miss DAG Automatic Cache Everywhere paid
Paid Boost plans extend the CLI with managed cache acceleration. Boost detects cacheable operations in your
build and test pipelines, surfaces cache hit rates in boost report,
and layers in acceleration so repeated agent workflows stay fast and predictable.
View how much you saved
Run boost report at any time to see total commands, tokens saved, average cache
hit rate, and the slowest commands by p90 latency.
$ boost report
Boost Report
══════════════════════════════════════════════════
Commands: 269 Tokens saved: 81.4K (70.4%)
Avg cache hit rate: 17.7%
Top Commands (by p90 duration)
Command Runs p90 Cache%
────────────────────────────────────────────────────
boost docker build 24 1.1s 17.7%
boost git add 5 136ms —
boost git status -u 2 98ms —
boost git commit 4 89ms —
boost git status 15 68ms —