CI Watchdog
GitHub App · dry-run by default

Stop paying for CI runs nobody needs.

Watchdog cancels superseded and stale GitHub Actions runs, flags runaways before they eat your budget, and explains every failed build in plain English. Safe by default, with a kill switch always in reach.

  • One GitHub App install
  • No workflow changes
  • Protected branches untouched
WATCHDOG · ACME-ORG watching
Minutes saved today
36
Runs watched
1,284
Mode
Live · 2 repos
  • RUNAWAY_DURATION

    acme/worker · test.yml · #990

    41 min elapsed, historical p95 is 12 min

    Alerted
  • STALE_ON_CLOSE

    acme/web · e2e.yml · #2207

    PR #318 merged 40s ago, 3 jobs still running

    Cancelled+22 min
  • SUPERSEDED

    acme/payments · ci.yml · #4812

    Newer SHA a91f3c2 already running on this PR

    Cancelled+14 min
Illustrative data · protected branches untouched · kill switch armed

Force-push storms

Every push starts a fresh run while the old ones keep billing, minute by minute.

Merged, still running

The PR is in. Its 20-minute e2e suite doesn't know and doesn't stop.

Silent runaways

A hung job sits until the 6-hour timeout. Nobody notices until the invoice.

How it works

Live in an afternoon. Trusted in two days.

Watchdog runs beside your CI, not inside it. Nothing changes in your workflows until you decide it should.

  1. 01

    Install the GitHub App

    Pick the repos. No workflow edits, no YAML in every repo. Webhooks start flowing in seconds.

  2. 02

    Watch it in dry-run

    For 48 hours Watchdog only logs what it would cancel, with the minutes and dollars attached.

  3. 03

    Go live per repo

    Flip repos live one at a time. First cancels can need approval, and the kill switch is always one click away.

The rules

Cancels what's obviously waste. Asks about the rest.

Rules are pure, tested functions. Only two ever cancel on their own. Everything else is a signal for a human.

Auto-cancel

Superseded runs

SUPERSEDED

A newer commit on the same branch or PR is already running. The older run is dead weight, so Watchdog stops it.

Auto-cancel

Stale on close

STALE_ON_CLOSE

The PR was merged or closed, but its CI keeps burning minutes. Watchdog cancels it the moment GitHub says so.

Alert first

Runaway duration

RUNAWAY_DURATION

A job runs far past its historical p95. You get an alert; it only cancels past a hard timeout you set.

Alert only

No relevant changes

NO_RELEVANT_CHANGES

Path filters say this workflow should not have run for the files that changed. Flagged for a human.

Fixer PR

Missing concurrency

MISSING_CONCURRENCY

Workflows with no concurrency block are listed in the digest, and Watchdog can open the PR that adds one.

Explains

Failure investigator

INVESTIGATION

Failed jobs get a structured diagnosis from an open-weight LLM, built from redacted, bounded log evidence.

Failure investigator

Red build? Get the why, not 4,000 lines of log.

Watchdog pulls a bounded, redacted slice of the failed job and asks an open-weight model for a structured diagnosis: what broke, the evidence, and the next safe step. It posts to Slack and the dashboard. It never runs a command.

acme/api · build.yml #1533
confidence 0.86
Summary

Dependency install failed: private registry rejected the token.

Evidence

npm ERR! 401 Unauthorized — GET https://npm.acme.dev/@acme%2fui

Next safe step

Rotate NPM_TOKEN in repo secrets, then re-run the failed job.

Illustrative example. Secrets redacted before the model sees anything.

Savings estimator

See what the waste costs you.

Drag the sliders to your numbers. Duplicate pushes and merged-PR runs add up faster than most teams expect.

Runner hours back each month
133
Saved per month
$48
Saved per year
$576

Linux ubuntu-latest at $0.006/min. Windows and macOS runners cost more, so savings scale up. Estimates, not invoice math. Watchdog shows your real numbers after 48 hours of dry-run.

Safety first

Built so you can say yes to auto-cancel.

A tool that cancels CI has to earn trust. Every guardrail is on by default.

Dry-run by default

Every repo is observed for 48 hours before a single real cancel.

Protected branches

main, master and release/* are never cancelled unless you opt in.

Global kill switch

Freeze every auto-cancel from the dashboard without a redeploy.

First-N approvals

Approve the first live cancels from Slack or the dashboard.

Explains every cancel

A PR comment says why it stopped and how to re-run.

No raw logs stored

The investigator sees redacted, bounded evidence only.

Slack alerts with approve / go-live / false-positive buttons Weekly digest of minutes saved
Plans

Start watching. Go live when you're sure.

One plan per GitHub App installation. Every plan starts in dry-run. Control pays for itself once it stops $39 of wasted CI a month.

Observe

$0 free forever

Dry-run only. Up to five repos. Alerts and the audit log, no live cancel.

  • Dry-run on up to 5 repos
  • Every rule, alert-only
  • Audit log
  • Weekly savings digest
Start observing

Control

Most teams

$39 / month

Go-live wizard, first-N approvals, Slack buttons, concurrency fixer PRs.

  • Unlimited repos
  • Live cancels + go-live wizard
  • First-N approvals
  • Slack action buttons
  • Concurrency fixer PRs
Choose Control

Org

$129 / month

Everything in Control plus invoice-backed usage and spend-spike alerts.

  • Everything in Control
  • Invoice-backed usage from GitHub billing
  • Spend-spike alerts
Choose Org
FAQ

Questions engineers ask first.

Will it ever cancel something on main?

Not unless you explicitly allow it. Protected branches (main, master, release/*) are excluded by default, on every plan.

Isn't GitHub's concurrency setting enough?

Only if every workflow in every repo has it, and most don't. Watchdog covers the gaps across the org, handles merged-PR waste concurrency can't, and opens PRs to add the block where it's missing.

What does the AI see from my logs?

A bounded, redacted excerpt of the failed job. Secrets and tokens are scrubbed first, raw logs are never stored, and findings are advice only. The AI never cancels or runs anything.

How accurate are the savings numbers?

They're estimates from elapsed time × jobs × runner rate. On the Org plan, Watchdog compares them with GitHub's own Actions billing data.

Can I self-host it?

Yes. It's a Node service plus SQLite, open on GitHub, with a Next.js dashboard.

Your next force-push storm is already queued.

Install Watchdog, watch it for 48 hours, and see exactly how many minutes it would have saved.