Disclosure: I work on Vigilhold, a hosted security agent for Linux servers. This post compares two open-source tools I don't make; the Vigilhold part is one paragraph at the end and clearly marked.
If you run a VPS with SSH open to the internet, your auth log fills with password guesses within minutes. The two tools most people reach for are fail2ban and CrowdSec. Both read your logs and block the addresses behind failed logins. They differ in three ways:
- where the decision is made
- where the block is applied
- whether you learn from other people's attackers
At a glance
| fail2ban | CrowdSec | |
|---|---|---|
| License | GPL-2.0, free | Engine MIT, free; optional Console with a free plan and paid plans |
| Reads | Log files (auth.log, sshd, Apache… out of the box) | Log files and HTTP requests from compatible web servers |
| Blocks with | The server's own firewall, for a set time | "Bouncers" for the firewall, web server, reverse proxy or CDN |
| Several servers | Each decides alone | Detect on one machine, block on many, via a local API |
| Shared intelligence | None | Community blocklist if you share what you detect |
| Web attacks | Whatever a log filter matches | AppSec (WAF) component with ModSecurity rules / OWASP CRS |
| Written in | Python | Go |
fail2ban in one paragraph
Filters (regexes) match log lines like a failed SSH password. When one address matches too often inside a time window, a jail runs an action, usually a firewall rule that rejects that address for a while. It's in every distro's repos and works the evening you install it. Its own README is honest about the limit: it lowers the rate of password guessing but can't fix weak authentication. Use SSH keys anyway.
CrowdSec in one paragraph
CrowdSec splits detection from enforcement. A log processor matches behaviour patterns ("scenarios"); separate remediation components ("bouncers") enforce decisions wherever you want: nftables/iptables, nginx, a reverse proxy, a CDN. A local API joins them, so one box can detect and a dozen can block. Share your detections and you get a curated community blocklist back, so addresses hammering other users can be blocked before they reach you.
Which one?
- One or two servers, key-only SSH, nothing else to babysit: fail2ban. Everything stays in text files on the box.
- Web apps, several servers, or blocking at the edge: CrowdSec, for the bouncers, the WAF component and the shared blocklist.
- They aren't exclusive. You can keep fail2ban for SSH and put CrowdSec in front of a web stack.
What neither one covers
Both tools are about who is knocking. Neither looks at what happens after a login succeeds, or after a vulnerable web app gets popped: a new process listening on a port, a cron entry that wasn't there yesterday, a web server making outbound connections. If that's the part that worries you, you need a host-level tool (AIDE, Wazuh, Falco, or a hosted agent), not a log banner.
Where Vigilhold fits (the disclosed part)
Vigilhold is the hosted agent I work on. It bans attacking addresses too, and also watches processes, network connections, file changes and logins, then responds only within limits you set, with every action in an audit log you can undo. It runs alongside fail2ban or CrowdSec and uses fail2ban for its own bans when fail2ban is installed. There's a longer version of this comparison with sources.
Sources (checked November 2026): fail2ban README · CrowdSec docs · CrowdSec on GitHub · CrowdSec Console pricing
What do you run on your servers, and why? I'm curious whether anyone's moved from one to the other.
Top comments (0)