The Sentinel blog
Notes on website monitoring, security practice, and the unglamorous work of keeping things online.
Sentinel now sends downtime, SSL and recovery alerts to Telegram. Add a chat, tap Start, and the Sentinel bot is connected: no bot to build, no token to paste. Groups, channels or direct messages, free on every plan.
How Sentinel added Telegram alerts in an afternoon without a toTelegram() on any notification: one channel class that renders the existing Slack message as Telegram HTML, anonymous routing for team chats, one bot for every tenant with a one-link connect flow, and the afterCommit trap in tests.
Internal observability tells you what your systems believe about themselves. External, multi-region monitoring tells you what customers actually get. Why black box checks from outside your network catch DNS, TLS, CDN and observability-stack failures that leave dashboards green, how heartbeat monitoring catches jobs that never ran, and a checklist to run against your current setup.
Sentinel's MCP server runs on Laravel's first-party MCP package, still pre-1.0, serving 26 tools to Claude and any MCP client. The engineering write-up: server anatomy, the bearer-vs-OAuth split, authorization as the real product, and the gotchas that only show up in production.
Agents can hold funded wallets now, so I catalogued what they can actually spend them on: a curated directory of live agent-payable APIs (x402 and MPP), organized by the capability gap each fills, with every listing continuously verified by a real uptime monitor.
Uptime alerts where your team already works: the official Sentinel Slack app is now listed on the Slack Marketplace, with per-channel installs, severity filtering, and a permission surface of exactly one capability.
PHP warnings above a WordPress header, stack traces from a failed deploy, database errors in the response: Sentinel now detects server-side errors automatically during normal HTTP monitoring, with no keywords to configure or maintain.
"My hosting company already monitors my site" is half true. The host watches its infrastructure, for its own SLA, from inside its own network. Expired certs, broken checkouts, DNS mistakes, and the question of who finds out first are all outside that scope.
Every incident starts with the same question: did we just ship? Deploy markers draw the answer directly on the response-time chart, cite deploys beside incidents, and can pause monitoring while a deployment runs.
Slack is a business signal and Discord is a hobbyist signal, so gating Discord monetized nobody. The reasoning behind the pricing change, and what the free plan includes now.
Set up your first monitor in about 60 seconds and let us do the watching.
Posts here are drafted with AI assistance and edited by me. The ideas, the decisions they describe, and every factual claim are mine.