Skip to content
Compile7Public

About

Paste a curl. CurlDog barks when it breaks, and tells you why. Self-hosted uptime monitoring on Cloudflare Workers.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

6 stars

Watchers

0 watching

Forks

Repository files navigation

CurlDog: Paste a curl. CurlDog barks when it breaks.

CurlDog

Paste a curl. CurlDog barks when it breaks, and tells you why.

Self-hosted uptime monitoring for teams, running entirely on your own Cloudflare account.
No servers, no containers, no database to babysit. Just a very loyal dog.

Deploy to Cloudflare

CI License: MIT PRs welcome Runs on Cloudflare Workers

Sponsored by SSOJet: enterprise SSO for B2B SaaS

Deploy Β· Features Β· How it works Β· Local dev Β· Contributing


CurlDog overview: one monitor down, project health, recent incidents



Sponsored by SSOJet: Enterprise SSO your security team will approve. Add enterprise SSO to your B2B SaaS in days, not months.


Why CurlDog

Most uptime tools make you rebuild your request in their form. Then, when something fails, they tell you that it failed, but not why.

CurlDog starts from the request you already have:

  1. Copy as cURL from your browser or terminal, and paste it in. Headers, auth, body and redirects all come along.
  2. Pick how often to check it.
  3. When it breaks, Google Chat or Telegram gets a message with the status, the response, an AI explanation of the likely cause, and buttons to Acknowledge or Mute it.

It all runs on your own Cloudflare account, behind your own login. Your API keys never leave infrastructure you control.

🐩 Meet Curly

Curly, the CurlDog mascot

Curly is an apricot poodle with one job: watch your endpoints and bark the moment one breaks.

  • Sniffs every curl on schedule, from up to 3 regions.
  • Barks in Google Chat, Telegram or Slack, with Acknowledge and Mute buttons.
  • Fetches the evidence: DNS, the redirect chain, and a diff against the last good response.
  • Keeps barking with reminders, and fetches a teammate if nobody comes.
  • Sits quietly when you mute a monitor for planned work.

The curly coat palette: Cream, Biscuit, Apricot, Caramel and Cocoa

✨ Features

Monitor anything you can curl

  • Paste any curl, including Chrome's Copy as cURL. See what it parses into, then Test it before saving.
  • Expected status (2xx, 200-299, 200,204), body contains, timeouts, redirects.
  • Multi-step checks: log in, save {{token}} from the JSON response, call the real endpoint with it.
  • Heartbeats for cron jobs and backups: they ping you, and you get alerted when they go quiet.
  • Check from 3 regions (US, EU, Asia). It's only marked down when 2 of 3 agree.

Alerts people actually read

  • Google Chat cards and Telegram messages with Open Β· Acknowledge Β· Mute 1h buttons. Slack, Discord, email, PagerDuty and SMS work too.
  • Each alert includes the status, a body snippet, 7-day uptime, and an AI summary of the likely cause (Workers AI).
  • One message per project when several monitors fail together, not a flood.
  • Reminders and escalation until someone acknowledges.
  • Slow alerts when responses stay over a threshold. Scheduled maintenance windows, or a quick Mute for deploys.

Diagnose without SSH-ing anywhere

  • The first try and the 3-second retry, side by side.
  • Redirect chain, DNS records (via 1.1.1.1), request ID, response size.
  • Diff against the last good response: which headers changed, and which body lines.

Built for a team

  • Projects group monitors. Notification channels are shared across projects.
  • Uptime for 24h, 7d and 30d, response-time chart, incident timeline.
  • Weekly report in your chat every Monday.
  • SSL certificate and domain expiry warnings at 14 and 3 days.
  • Public status page per project: uptime bars, incident updates you write, RSS feed, JSON and a badge.
  • Monitors as code: pull, plan and apply a JSON file over the API.
  • Import from Uptime Kuma, one project per site, notifications included.
A failing monitor with an AI explanation and acknowledgement Diagnose view comparing the failed response with the last good one
An incident: what failed, the likely cause, and who's on it Diagnose: both attempts, DNS, and a diff against the last good response
Monitor form with a login step that saves a token Project page in dark mode with sparklines
Multi-step checks: log in, then reuse {{token}} Dark mode, mobile layout, no framework

πŸš€ Deploy in 5 minutes

1. Click the button

Deploy to Cloudflare

Cloudflare copies this repo to your GitHub account, creates the D1 database, applies the schema, and deploys. Every push to your copy redeploys it.

Prefer the command line?
git clone https://github.com/Compile7/curldog.git
cd curldog
npm install
npx wrangler login
npm run deploy

npm run deploy creates the database on first run, deploys the Worker, and applies migrations.

2. Lock it down with Cloudflare Access

Open your new Worker's URL. You'll see a setup page, because CurlDog refuses every API request until Cloudflare Access protects it. Nothing is exposed while you set it up.

  1. In the Cloudflare dashboard: Workers & Pages β†’ curldog β†’ Settings β†’ Domains & Routes. Next to your workers.dev URL, choose Enable Cloudflare Access. (On a custom domain, add a self-hosted Access application for it instead.)

  2. Manage Cloudflare Access β†’ edit the policy to allow your team, for example emails ending in @yourcompany.com.

  3. From that Access application, copy the Application Audience (AUD) tag and your team domain (like acme.cloudflareaccess.com).

  4. Settings β†’ Variables and Secrets β†’ Add:

    Name Value
    ACCESS_TEAM_DOMAIN acme.cloudflareaccess.com
    ACCESS_AUD your AUD tag
    APP_URL (optional) e.g. https://status.acme.com. Links in alerts use it. Without it, CurlDog uses the address you open it at.
  5. Reload. You'll sign in through Access and land on an empty dashboard.

Variables you set in the dashboard survive later deploys (keep_vars is on).

3. Connect your chat

Go to Notifications, pick a channel, and click Send test.

Channel Setup
Google Chat Space β†’ Apps & integrations β†’ Webhooks β†’ add one, paste the URL.
Telegram Message @BotFather β†’ /newbot β†’ paste the token. Add the bot to your group, send a message, click Detect.
Slack / Discord Paste an incoming webhook URL.
Email Create a Resend API key and verify your sending domain. Free for 3,000 emails a month.
PagerDuty Service β†’ Integrations β†’ Events API V2 β†’ paste the integration key. Down opens an incident, Acknowledge acknowledges it, and recovery resolves it.
SMS Paste your Twilio Account SID, Auth Token and sender number. Only incidents are texted, not reports.

4. (Optional) Open the public paths

Two paths are meant for people and servers that can't sign in to Access:

  • /hb for heartbeat pings from your cron jobs. Each ping URL carries its own random token, and a ping can only say "I'm alive".
  • /status for public status pages. They show the monitors you pick and the incidents you post, never URLs or errors.

For each one you use, add an Access application for that path on the same domain, with a Bypass policy for Everyone.

🧭 Using it

Public status page

In Project settings, tick Public status page, pick an address like /status/acme, and tick the monitors to show. To show a friendlier name ("Checkout" instead of "POST /v2/checkout"), edit the monitor and open Status page.

The page shows:

  • an overall banner: operational, degraded (slow), partial or major outage, or maintenance,
  • one line per monitor, with 30 days of daily uptime bars,
  • incidents you posted, with every update,
  • open and upcoming maintenance,
  • outages the monitors caught on their own, from the last 14 days.

Post an incident with the megaphone button on the project: a title, minor or major impact, and a message. Add updates as you go (Investigating, Identified, Monitoring, Resolved). An open incident sets the banner, even if the monitors look fine.

Next to the page:

URL What it's for
/status/acme/feed.atom RSS/Atom feed. Subscribe in a feed reader, or in Slack with /feed subscribe.
/status/acme/status.json Current status, shaped like Atlassian Statuspage's status.json, so existing tools can read it.
/status/acme/badge.svg A badge for your README: ![status](https://…/status/acme/badge.svg)

It never shows URLs, error messages, responses or who posted. Everything is cached for a minute.

Maintenance windows

On a project, click the wrench and schedule a window: once, every day, or every week. While it's open:

  • the project's monitors aren't checked, so nothing alerts and no downtime is counted,
  • the status page says "Scheduled maintenance in progress".

For a quick one-off, Mute still works too. Mute keeps checking but stops the messages.

API and monitors as code

Everything the UI does goes through a JSON API under /api. Scripts sign in with a Cloudflare Access service token:

  1. Zero Trust β†’ Access β†’ Service credentials β†’ create a service token.
  2. In CurlDog's Access application, add a policy with action Service Auth that includes that token.
  3. Send CF-Access-Client-Id and CF-Access-Client-Secret headers with each request.

scripts/curldog.mjs keeps your monitors in a JSON file you can review in pull requests:

export CURLDOG_URL=https://curldog.acme.workers.dev CF_ACCESS_CLIENT_ID=… CF_ACCESS_CLIENT_SECRET=…
node scripts/curldog.mjs pull > curldog.json   # export what's live
node scripts/curldog.mjs plan curldog.json     # show what would change
node scripts/curldog.mjs apply curldog.json    # create and update; add --prune to delete monitors not in the file
  • Projects and monitors are matched by name. Fields you leave out keep their current value.
  • Channels are referenced by name and never created from the file, because they hold secrets.
  • ${NAME} in the file is replaced with the environment variable NAME, so API keys in curl headers can stay out of git.
Multi-step check: log in, then call an authenticated endpoint

Under Steps before this request, add a step:

curl https://auth.example.com/oauth/token --json '{"client_id":"…","client_secret":"…"}'

with the variable token = $.access_token. Then make the main request:

curl https://api.example.com/v1/me -H 'Authorization: Bearer {{token}}'

Variables can come from the JSON body ($.data.items[0].id) or a header (header:Set-Cookie). You can have up to 5 steps.

Heartbeat: get alerted when a cron job stops running

Create a monitor, choose Heartbeat, and pick how often the job runs. Add the ping URL to the end of the job:

0 2 * * * /usr/local/bin/backup.sh && curl -fsS -m 10 --retry 3 https://curldog.example.com/hb/3f9c…

If no ping arrives within the interval plus half an interval of grace (at least a minute), the monitor goes down.

Import from Uptime Kuma

In Uptime Kuma: Settings β†’ Backup β†’ Export. In CurlDog, click Import and drop the JSON file.

  • HTTP and keyword monitors come over, with headers, body, basic auth, status codes and intervals.
  • One project per site, named after the brand ("api.ssojet.com" and "ssojet.com" both go into SSOJet).
  • Google Chat, Telegram, Slack and Discord notifications become channels.
  • Re-importing is safe: duplicates are skipped.
What "down" means
  • A failing check is retried once after 3 seconds. A monitor goes down only if both attempts fail.
  • With 3 regions on, there's no retry. It's down when at least 2 of the 3 regions fail.
  • Alerts fire when a monitor goes down and when it recovers, not on every failed check. Reminders and escalation are set per project.

πŸ— How it works

flowchart LR
  U[Your team] -->|Cloudflare Access| W
  subgraph CF[Your Cloudflare account]
    W[Worker<br/>API + UI] --- D[(D1<br/>monitors, checks, incidents)]
    C[Cron Β· every minute] --> W
    W -->|opt-in| P[Durable Objects<br/>US Β· EU Β· Asia]
    W -->|once per incident| AI[Workers AI]
  end
  W -->|curl requests| T[Your endpoints]
  P --> T
  J[Your cron jobs] -->|/hb/token| W
  W --> G[Google Chat Β· Telegram Β· Slack<br/>Discord Β· Email Β· PagerDuty Β· SMS]
  V[Your customers] -->|/status/slug| W
Loading
  • One Worker serves the UI (static files, no build step) and the JSON API.
  • Cron runs every minute. It claims due monitors with a single UPDATE … RETURNING, so overlapping runs never double-check. It runs them 50 at a time, records results, and sends alerts grouped by project.
  • Every hour it cleans up old data (checks after 30 days, incidents after 90), refreshes SSL and domain expiry, and on Mondays sends the weekly report.
  • Auth is Cloudflare Access. The Worker verifies the Access JWT on every API call itself (signature, audience, issuer, expiry), so going around Access doesn't work.

πŸ’Έ What it costs

Free plan Workers Paid ($5/month)
Good for Trying it out, about 20 monitors on 5-minute checks Hundreds of monitors on 1-minute checks
Limit you'll hit first D1 writes (100k rows/day) and 50 outbound requests per cron run Rarely anything
AI summaries Free tier covers it: one call per incident Same
3-region checks Off by default, turn on per monitor A few cents per monitor per month

There's no screenshot or browser rendering, which keeps it cheap.

πŸ›  Local development

npm install
npm run dev        # http://localhost:8787, signed in automatically on localhost
npm run seed       # optional: demo projects and 24h of history (replaces local data)
npm run check      # type-check and tests

Cron doesn't run on its own in dev. Trigger it with:

curl "http://localhost:8787/cdn-cgi/handler/scheduled"

Local dev runs with --local, so AI summaries are skipped (Workers AI only runs on Cloudflare). See CONTRIBUTING.md for the code layout and house rules.

⚠️ Good to know

  • Public URLs only. Checks run from Cloudflare's network. Reaching private services would need Cloudflare Tunnel.
  • Not supported: -k/--insecure (Workers always verify TLS), --resolve, proxies, client certificates, -F multipart, @file bodies. The form tells you why.
  • Secrets are stored in plain text in D1. Anyone who can sign in can see curl headers and channel tokens. Give access to people you'd trust with them.
  • SSL expiry comes from Certificate Transparency logs (crt.sh), because Workers can't read a server's certificate directly. Domain expiry comes from RDAP; some registries (.ai, for example) don't publish it.
  • Past roughly 1,000 monitors per minute, checks should move onto a Cloudflare Queue. It's on the roadmap.

πŸ—Ί Roadmap

  • Public status pages per project, with incident updates, feed, JSON and badge
  • PagerDuty, email and SMS channels
  • Scheduled maintenance windows
  • API and monitors as code
  • Microsoft Teams channel
  • Custom domains and branding for status pages
  • Email subscribers for status pages
  • 90-day status history (needs daily rollups; checks are kept 30 days)
  • JSON assertions ($.status == "ok")
  • TCP and DNS monitors
  • Queue-based fan-out for very large installs

Have an idea? Open a feature request.

🀝 Contributing

Contributions of all sizes are welcome: a typo fix, a new curl flag, a new alert channel. Start with CONTRIBUTING.md. npm run seed gives you a populated app in seconds.

Please follow the Code of Conduct. Found a security issue? See SECURITY.md and report it privately.

πŸ’œ Sponsor

CurlDog is sponsored by SSOJet. SSOJet adds enterprise SSO to your B2B SaaS in days, not months. Connect 100+ identity providers over SAML, OIDC and SCIM without rebuilding your auth.

πŸ“„ License

MIT Β© Govind Malviya and contributors.

About

Paste a curl. CurlDog barks when it breaks, and tells you why. Self-hosted uptime monitoring on Cloudflare Workers.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

6 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages