Start freeStart free
3.4Flaky tests now quarantine themselves →

CI that only runs what your diff touched.

Runnel reads each pull request, walks your dependency graph and runs only the builds and tests that could have changed. Everything else comes from cache.

GitHub · GitLab · BitbucketReads Nx, Turborepo, Bazel, Gradle and Cargo graphsYour runners or ours

Example: pull request 4182 changes three files in the money-core package. Runnel finds four affected packages out of fourteen, schedules 212 of 1,840 tests, pulls the remaining builds from cache, and passes in 1 minute 52 seconds compared with 14 minutes 8 seconds for a full run.

(Product)

Skipping work is only safe if it’s exact.

01 — REMOTE CACHE

Every build is computed once, for the whole team.

Runnel hashes inputs, not timestamps: source, lockfile, toolchain and env. Same hash, same artefact, pulled in under a second by every runner and laptop.

  • Content-addressed SHA-256 keys
  • Retention 30 days, or pin forever
  • Local hit ~300 ms
Cache · last 14 daysatlas
94.2% hit rate 412 GB served · 18,630 builds skipped
Oct 4Oct 10Today
Most reusedHitsSizeLast
ui-kit:build/dist4,11238 MB2m
config:codegen3,8802 MB2m
auth:build/dist2,96411 MB6m
02 — FLAKY-TEST QUARANTINE

A test that flips three times stops blocking merges.

Runnel watches every result. When a test passes and fails on the same code, it’s quarantined: it still runs and reports, but it can’t turn a PR red. You get a ticket, not a re-run button.

  • Threshold 3 flips in 20 runs
  • Release automatic after 50 clean runs
Flaky tests · 2 quarantinedlast 20 runs
03 — RUNNEL REPRO

Any red task, on your laptop, in one command.

runnel repro pulls the exact inputs, cache and env of a CI task and replays it locally. No more “works on my machine” in review threads.

  • Pulls remote cache + env
  • Install macOS, Linux, WSL
~/atlas — zsh
A water channel cut into dark basalt splits into two, lit by a single amber lamp

One change, two paths. Runnel only follows the one your diff took.

(Your numbers)

Don’t take our word. Run it on your repo.

$ runnel estimate --last 200example output
Merged PRs, by sizePRsFullRunnel
1 package touched12114m 08s1m 41s
2–3 packages5414m 08s3m 56s
4+ packages1914m 08s9m 12s
Lockfile / toolchain change614m 08s14m 08s
Median PR20014m 08s2m 03s

Dry run against your last 200 merged PRs. Reads the graph and history, runs nothing, needs no account. Output above is from a 14-package demo monorepo; yours will differ.

40,000
70%
Estimated minutes back each month 28,000

At $0.008 a runner-minute, that’s about $224 a month, before anyone stops waiting.

(Pricing)

Pay per seat. Skipped minutes are free.

Hobby

$0forever
1 seat

For side projects and open source. The whole product, smaller limits.

  • 2,000 runner-minutes / month
  • 5 GB remote cache
  • Affected-only runs and repro
  • Community support
Start free, no card needed

Team MOST TEAMS

$22per seat / month
billed monthly

For product teams shipping a monorepo every day.

  • 25,000 runner-minutes included
  • 200 GB cache, 30-day retention
  • Flaky-test quarantine
  • Google and GitHub SSO
Start a 14-day Team trialNo card for 14 days

Enterprise

Custom

For platform teams with their own runners, rules and auditors.

  • Self-hosted runners and cache
  • SAML, SCIM and audit log
  • 99.95% uptime SLA
  • A named engineer, not a queue
Book a 30-minute call

Extra runner-minutes are $0.008 each. Cache hits and skipped tasks never count as minutes.

(Start)

Stop waiting on code you didn’t touch.

Invite

Check your inbox.