Before that, I used WUD, but at the time Versentry was created, it didn’t have a proxy for Telegram notifications, which was the impetus. As a result, I ended up with a micro‑service that fully meets my needs.

What it does: reads running containers from the local Docker socket, compares their tags and digests against OCI registries, and sends a notification when something's newer. It never pulls images or touches containers. That's the whole point - notify-only by design.

A few things that set it apart:

• Central tag-filter rules in config (regex by image repo), so you're not forced to label every container. Per-container versentry.include labels also work if you prefer them.
• Built-in proxy (SOCKS5/HTTP) for notifier delivery and registry traffic - handy behind network restrictions.
• Any OCI registry - Docker Hub, GHCR, Quay, GitLab out of the box; private/self-hosted via type: oci.
• Notifiers: stdout, Telegram, Discord, Gotify, ntfy, and a generic webhook.
• Readable notifications out of the box - sensible default messages, no template-wrangling required; Go text/template overrides are there if you want to customize.
• semver/numeric detection (cross-major updates get flagged) plus digest comparison for floating tags like latest.
• Tiny ~5 MB scratch-based image, MIT licensed.

It's probably not for you if you need: a web UI (try WUD), automatic pull-and-restart (that's Watchtower's job), non-Docker providers like Kubernetes/Podman (Diun covers more), or Slack/email as first-class notifiers.

It's a young project - fewer battle-tested deployments than the established tools - so I'd genuinely welcome feedback from anyone who runs it.

GitHub: https://github.com/BlackRaincoat/versentry

AI Disclosure:

• Design - Pair
• Implementation - Generated
• Testing - Assisted
• Documentation - Generated
• Review - Assisted
• Deployment - Generated

Note: The idea, the architecture decisions, and all real-world testing were mine - I ran it across multiple live servers and that's what caught the actual bugs. The AI wrote the implementation, the test code, docs, and CI config from my prompts. I reviewed behavior on real deployments rather than reading every line, so code-level review leaned on the AI.