Ceryx 1.0

We are truly thrilled to announce the general availability of Ceryx; our open source programmable network gateway and one of the most (if not the most!) robust pieces of software we have ever written.

Ceryx is the front-door for all of our clusters (production, preview environments, internal tooling, Fakos tenant cells, everything). It's reliable and because it's based on NGINX it can take extreme loads of traffic and forward it to the appropriate container (or micro-VM with Fakos lately!) with just a sliver of resources.

Besides proxying traffic, Ceryx also takes care of issuing and renewing the HTTPS certificate of each route automatically. The last piece is how effortless it is to publish tens of new routes each day via the Ceryx API in GitHub Actions, as we always deploy new preview environments for the work we do on our clients' and our own products. No configuration file to maintain. No server restarts. It can handle millions of active routes without a flinch. This makes it also the perfect solution for agents, which spin up new ephemeral environments at incredible speeds.

A fun part about the story of Ceryx is that it dates back more than ten years. We started working on it and released the first open source code during our time in our first startup, the online IDE SourceLair, where we used it to publish the web server of each project at its own unique URL. It has gone through multiple rewrites, from Python implementations with Twisted, then Tornado, then NGINX with scripted reloads, then OpenResty with Redis backend and a Python API. And it has been battle-tested for more than a decade in production, routing traffic to literally millions of containers and back-end services.

Without diving too much into the technical details at this moment, we can confidently say that we have reached the architecture we are most happy with and this is what drove the v1.0 release. Ceryx runs in a single container. Its API and certificate issuing are implemented directly with LuaJIT inside NGINX. Finally, the data store is memory-mapped SQLite to ensure direct memory access with rapid speeds when picking a route. We will start discussing more technical details in the future as we move forward with further development of Ceryx.

We have big plans for Ceryx, which fit in a greater picture of how we believe software can be built in this exciting new age we live, including but not limited to AI agents.

Until then, give it a go at https://ceryx.dev and stay tuned!