hone
⌘KOpen app →
verb · to sharpen

Hone runs only the
tests that matter.

We predict which tests your PR actually needs, skip the rest, and post a single honest comment with what we saved — and how we got there.

TIME SAVED · NOV 2025
14h 32m
+ €71.16 in CI compute
ACCURACY
96.1%
measured by full-suite shadow runs
FREE CI AUDIT — NO SIGNUP
Tell us your repo. We'll guess what hone would save you. Honest numbers — we show how we got there.
ⓘ Demo build — no repo data leaves the browser.
HhoneHoned2m
SAVED
4m 12s
full suite ran in background · 0 missed failures
Files touched: 3 · related tests selected via dependency graph + history.
how it works

Three pieces. Four minutes to install. Zero YAML.

01
Watch in shadow

We install as a GitHub App, watch the full suite for 14 days, and post the saving we *would* have made — without changing your CI.

02
Comment on every PR

One comment, hero number, files touched, the test list, and exactly why each test was selected — or skipped.

03
Switch to active

When you're convinced, flip on with a rollout dial (25% → 100%). The full suite stays on as a safety net for 7 days.

the most-seen surface

Three states. Same shape. Honest about every one.

Shadow shows what we *would* skip. Honed shows what we did skip. Miss is the trust-builder — when we're wrong, we say so on the same comment.

ShadowHonedMiss
HhoneShadow2m
WOULD HAVE SAVED
4m 12s
142 / 1,847 tests would run · day 8 of 14
Files touched: 3 · related tests selected via dependency graph + history.
HhoneHoned2m
SAVED
4m 12s
full suite ran in background · 0 missed failures
Files touched: 3 · related tests selected via dependency graph + history.
HhoneMiss2m
WE MISSED ONE
billing.spec.ts
full suite caught it · 1st miss in 142 PRs
The dependency from /lib/tax.ts wasn't in our graph. We've added it.
numbers
Avg time saved per PR
4m 12s
Misses in last 1k PRs
9
Repos in shadow
412
CI compute saved · Q4
€18,402
ask hone — ai demo

The docs you can talk to.

Streaming via the Vercel AI Gateway. Drop in your own model in apps/web/app/api/chat/route.ts — we route through 'anthropic/claude-sonnet-4-6' by default.

ASK HONE — AI DEMO
Ask anything about how hone selects tests, handles misses, or rolls out. Powered by Claude via the Vercel AI Gateway.
start in shadow · no commit changes

Watch hone for 14 days. Pay only when it earns it.

Run the free audit →