open-appsec

Requests scored by two learned models inside your existing proxy

open-appsec is an open source web application firewall (WAF) from Check Point Software Technologies that also covers APIs. Its engine does not match requests against signatures. A first model, trained offline on millions of malicious and benign requests, scores each decoded request. When that model flags one as suspicious, a second model, built from the protected site's own traffic, weighs the URL and the users involved to produce the confidence score that decides whether to block.

open-appsec sits inside the reverse proxy you already run instead of replacing it. A C module, the "attachment", plugs into NGINX, Kong, APISIX or Envoy and hands requests to a C++ agent over shared memory. BunkerWeb takes the opposite route: it becomes the reverse proxy and filters with the OWASP Core Rule Set. CrowdSec starts from logs and gets IP addresses blocked.

The open source WAF comparison weighs this learned-model approach against OWASP rules and community reputation.

Staged learning in Learn/Detect before Prevent mode

• No signature feed to keep current: the documentation promises zero-day protection with no signature updates, and none of the exception handling that rule-based WAFs need.
• Learning in stages: Learn/Detect mode logs without blocking. With enough traffic, learning takes "about 2-3 days", and the docs recommend switching to Prevent once the engine reaches "Graduate", the third of five levels.
• Works with the proxy you have: NGINX, Ingress NGINX, Kong, APISIX, Envoy, Envoy Gateway and the Istio ingress gateway, on Linux, Docker or Kubernetes. On Linux, the installer picks a precompiled attachment for your server version; if yours is not on the list, you build it from source.
• NGINX Proxy Manager and NPMplus: the team ships NPM images with the attachment built in, so the WAF is configured from the NPM interface with no SaaS involved.
• CrowdSec in both directions: open-appsec can act as a CrowdSec bouncer and block addresses from CrowdSec's threat intelligence (Ingress NGINX on Kubernetes, or Docker with NGINX or Kong). The other way round, the openappsec/openappsec collection lets CrowdSec parse open-appsec's logs and feed them into its community blocklist.
• Local or hosted configuration: a local_policy.yaml file, Kubernetes CRDs, a Helm chart, or the web UI at my.openappsec.io.

Production model behind a login, extra engines on paid plans

The model in the repository is not the production one. The supervised model shipped with the code, called "basic", is Apache-2.0 like the rest of the repository, and the README recommends it for test and monitor-only setups. The "advanced" model, recommended for production, is downloaded from the my.openappsec.io portal after logging in, under its own "Machine Learning Model license".

The extra engines are paid. Anti-bot, API schema enforcement, file security and IPS, which brings signatures back (more than 2,800 web CVEs), require a subscription to Check Point WAF (the enterprise edition) or to the older Premium Edition, which the docs mark "deprecated". In the community edition, rate limiting accepts a single rule. The pricing page still sells Premium Edition at $79 to $109 a month, next to an offer "extended through Dec. 31, 2024".

Running without the SaaS takes six containers, PostgreSQL included. The repository's NGINX Compose file starts the agent and an NGINX container carrying the attachment. Without a portal token, the standalone profile adds smartsync, which merges learning across instances, its storage service, the tuning service and PostgreSQL 18. The agent shares the host's IPC namespace (ipc: host), which the README makes mandatory. The install pages give no memory or CPU figures, and the example pulls the latest image.

Code lands in batches. The main branch has 10 commits in 2026, one of them a 216-file drop titled "Jan 06 2026 dev". Four releases shipped this year, up to 1.1.36 in late August, against ten in 2025. The notes for the last three come down to "Support for new nginx & Linux distributions" and "General bug fixes".

The audit dates from launch. The README points to an independent third-party code audit carried out in September and October 2022, and Lexfo's report sits in the repository. The OpenSSF Best Practices badge, at "passing" level, was earned on October 26, 2022, the day the repository was created, and its entry has not been updated since.