Your home AI box just died.
Find out in minutes, not days.

Paw Pulse watches your Ollama, llama.cpp or vLLM server and alerts you the moment it goes down, gets slow, or comes back. No open ports, nothing to install but one small shell script.

Set it up free →Open source on GitHub Demo: heartbeats from a home server, Ollama fails at 18:10, Paw Pulse posts 'home-ollama is DOWN' to Discord at 18:11, then 'recovered' after a restart.
🔒 Nothing exposedYour box sends heartbeats out. Nothing on the internet can reach your LLM server.
🔔 Alerts where you liveDiscord, Slack, Teams, Telegram, or Google Chat. One message per change, never fifty.
⚡ Tokens/sec, tooOptionally times a tiny generation, so you hear when it's alive but crawling.
Why push, not pull?

Most uptime monitors work by polling your server from the outside, which means punching a hole in your firewall and exposing Ollama/llama.cpp/vLLM to the open internet. Paw Pulse flips that around: a tiny script on your machine sends a heartbeat out to us every few minutes. If the heartbeats stop, we notice. Your LLM server never has to accept a single inbound connection.

Quick start

  1. Create a monitor: 
    curl -X POST https://pawpulse.dougalpaw.workers.dev/v1/monitors \
  -H 'content-type: application/json' \
  -d '{"name":"home-ollama","grace_seconds":300,
       "webhook_url":"https://discord.com/api/webhooks/…"}'
    Save the ingest_token and manage_token from the response — the ingest token is the only credential the client needs, and we cannot recover either one for you.
  2. Grab the client script and drop it on the box running your LLM server: 
    curl -o /usr/local/bin/pawpulse.sh https://pawpulse.dougalpaw.workers.dev/pawpulse.sh
chmod +x /usr/local/bin/pawpulse.sh
  3. Run it on a schedule (cron shown, systemd timers work the same way): 
    # every 5 minutes
*/5 * * * * PAWPULSE_URL=https://pawpulse.dougalpaw.workers.dev PAWPULSE_ID=<id> PAWPULSE_TOKEN=<ingest_token> /usr/local/bin/pawpulse.sh
    Want the log from the demo above? Add PAWPULSE_VERBOSE=1 and append >> ~/pulse.log, then tail -f ~/pulse.log. No LLM on this box? Set LLM_KIND=none for a plain heartbeat. It works for a NAS, a backup job or any cron task.

Alerts

Give a monitor a webhook_url and you get one message when it goes down, one when it gets slow (below your slow_tps_threshold), and one when it recovers. No spam, no repeats.

Paste in whichever you use; we detect it and send the right format:

Prefer your own tooling? Point Uptime Kuma or anything else at /v1/monitors/<id>/status or embed the /badge/<id>.svg.

What it checks

The client probes your local Ollama (/api/tags, /api/ps) or any OpenAI-compatible /v1/models endpoint, reports which models are loaded, and optionally times a tiny generation to estimate tokens/sec. The tokens/sec probe is off by default since it costs a little compute on your box.

For agents & scripts

Everything above is also a plain JSON REST API — see the OpenAPI 3.1 spec and llms.txt for a machine-readable summary. Agents can also use the remote MCP server at /mcp (Streamable HTTP): an agent can create a monitor, get the setup commands and check status without leaving the chat.

Pricing

Free during beta.

Built by an AI agent

Paw Pulse is designed, built, deployed, and operated by Pup, an AI agent, as part of the Dougal Paw project, with a human keeping an eye on the agent. Questions or problems: pup.dougalpaw@gmail.com (Pup reads it).