Project Outerlude - A Modern Interlude aCis fork

[Elfo]

I came out of my cave as I do once every 5 years. By now, I know nobody really cares about L2, but I still find it fun to experiment.

Everything you see here will be free and open source. I have no interest in selling anything.

 

Long story short, I like to revisit Interlude and apply what I've learned to see how far I can push it. Here's Outerlude, a public fork of aCis for the modern age.

 

Video demo:

 

Work that has been done:

 

• Redone the netcode from scratch to be async
• The NPC AI was completely redone based on Finite State Machines
• Moved to PostgreSQL and using some of its cool features
• Lots of config that should be hot reloadable has moved to the database
• OpenTelemetry instrumentation, where it makes sense, and a Grafana dashboard
• A built-in REST API for server management
• A built-in MCP Server for LLMs
• Nidrah AI, an AI Agent to make managing the server easier
• Real-time server map view
• Chat auditing and live snooping
• A new Fake Players Engine with a Node logic system and a new LLM planner for any behavior
• Just watch the video

 

If there is interest in this and I'm happy with it, or I get bored (which I always do), I will open-source it.

Let me know what you think and if there is some feature you'd like me to implement.

[Splicho]

3 hours ago, Elfo said:

• Moved to PostgreSQL and using some of its cool features
• A built-in REST API for server management

OH MY LORDDDDDDDDDDDDDDDDDDDDD

 

FINALLY

Edited May 13 by Splicho

[TrakinasXP]

Will this CodeIgniter-based web panel be made available?

In any case, we'll contribute.

[aguy]

LLMs are getting cheaper. Especially the small models are cheap and very capable, very soon we will see AI agents per fake player in games. Imagine 500 AI agents having access to various functions (MCP) and going for pvp, farming, setting up shops, recruiting, writting memories (pgvector), reading memories, taking decisions against other players, pathfinding, organizing sieges ... 

[Nightw0lf]

Migrating a legacy Interlude server to PostgreSQL while adding real observability is basically forcing 2006 MMO engineering to attend a 2026 infrastructure conference at gunpoint.

PS: which revision of aCis?
PS: what was broken during this whatever you call it. 

 

AAC Guard beign asked to adapt to this be like:

- Creating bugs since early 2018

Edited May 14 by Nightw0lf

[Seamless]

First of all, its great to finally have something meaningful to discuss on this forum  and good job pushing this community forward, even if its mostly dead. I always wondered why, in the last 15 years, no one has properly documented any of the source code (acis mobius) or at least exposed common functionality through something like a rest api, which is straightforward enough for most people to understand and use.

 

I have some technical questions since im really interested

 

Regarding real time features like maps and enchanting  are you using technologies like grpc? And what happens when both a logged in player and panel try to enchant the same item at the same time? Do we run into a race condition how do you handle that ?

 

Regarding  fake player system  external models calls typically take 1-2 seconds to respond, while an local models might take around 200-300ms. During that time a lot can happen in game. So when the plan becomes stale you override it and wait for the new one  meaning the node handles everything in real time and the LLM simply sets the goal asynchronously. Is that the correct understanding ?

 

Regarding the security aspect a lot can go wrong when exposing crucial server logic to the open web (experimental is good). My real concern is  the mcp in the panel where  a lot can go wrong with bad user input. 

 

16 hours ago, Elfo said:

I like to revisit Interlude and apply what I've learned to see how far I can push it

 

Wasn't the game client always the limiting factor for l2j?

[Frank]

Amazing job! Gratz!

[Elfo]

13 minutes ago, Seamless said:

Regarding real time features like maps and enchanting  are you using technologies like grpc? And what happens when both a logged in player and panel try to enchant the same item at the same time? Do we run into a race condition how do you handle that ?

There is no need for gRPC in this case, even tho originally it was gRPC based but since we don't need it to be bi-directional, we switched to simple http requests for the web calls and SSEs for the data streamed from the server. There are distributed locks in place to precent race conditions between actions that can happen between multiple web instances and the server.

 

13 minutes ago, Seamless said:

Regarding  fake player system  external models calls typically take 1-2 seconds to respond, while an local models might take around 200-300ms. During that time a lot can happen in game. So when the plan becomes stale you override it and wait for the new one  meaning the node handles everything in real time and the LLM simply sets the goal asynchronously. Is that the correct understanding ?

Local models can also be slow depending on the model, and most external models can actually be faster than local ones if you use Flash 2.5 or something along those lines. I am running on 512GB of Unified Memory on my Mac Studio M3 Ultra so the speed of the local model for a small model is pretty good but I tested it with Gemini too and it works equally as fast and in some cases faster. The way it works is that I'm using pgvector (one of the benefits of moving to Postgres) to search the data and see what the player can see etc and there is some batching of the next few actions for 2-4 seconds for the user until the next LLM request fires. The batching also includes branching on logic so if they for example fall under some HP they will move to kiting instead of attacking or maybe they heal etc.

 

13 minutes ago, Seamless said:

Regarding the security aspect a lot can go wrong when exposing crucial server logic to the open web (experimental is good). My real concern is  the mcp in the panel where  a lot can go wrong with bad user input. 

Everything is authed and permission-based. The server and the backend of the frontend have secure communication between them, either with a symmetric key (not recommended for production) or a certificate (the recommended way), so there is no worry. It's all tied to the account's access level, etc., so nobody can make an action that they normally wouldn't be allowed to do. Even the MCP is token-based, and there are prompt injection protections in place. The MCP is audited, and every mutation needs confirmation. The admin area is only accessible to the admin account anyway so normal users can't access it.
 

Edited May 14 by Elfo

[Elfo]

The account control panel also comes with a real-time web-based L2 Client so you can see what's going on in the server without needing to load the client:
 

 

[Litch]

2 hours ago, Elfo said:

The account control panel also comes with a real-time web-based L2 Client so you can see what's going on in the server without needing to load the client:
 

 

this is isane bro!

[Elfo]

Textures and animations also work

 

[Seamless]

@Elfo Thanks for your reply. At this point, I’m more interested in your ai  workflow rather than the actual implementation. Even if you don’t open source it, it would still be really valuable for research if you could share the technologies you used in each step. Great job 

 

On 5/14/2026 at 1:40 PM, Elfo said:

I am running on 512GB of Unified Memory on my Mac Studio M3 Ultra

 

Holy cow thats about 15k

 

On 5/18/2026 at 8:11 PM, Elfo said:

The account control panel also comes with a real-time web-based L2 Client so you can see what's going on in the server without needing to load the client:

Webgl ?

Edited Tuesday at 07:06 PM by Seamless

[Elfo]

1 hour ago, Seamless said:

Holy cow thats about 15k

14,299 GBP


 

 

1 hour ago, Seamless said:

Webgl ?

Yeah and it loads the actual files from the actual client so it's the same unr, utx, etc files that you have on the client.

[Alicio]

On 13/5/2026 at 13:30, Elfo said:

Salí de mi cueva como lo hago una vez cada 5 años. A estas alturas, sé que a nadie le importa realmente L2, pero todavía me resulta divertido experimentar.

Todo lo que veas aquí será gratuito y de código abierto. No tengo ningún interés en vender nada.

 

En pocas palabras, me gusta volver a visitar Interlude y aplicar lo que he aprendido para ver hasta dónde puedo llevarlo. Aquí está Outerlude, una bifurcación pública de aCis para la era moderna.

 

Demostración en vídeo:

 

Trabajo que se ha realizado:

 

• Rehizo el código de red desde cero para que sea asíncrono
• La IA del NPC fue completamente rehecha basándose en máquinas de estados finitos
• Se trasladó a PostgreSQL y utiliza algunas de sus funciones interesantes
• Muchas configuraciones que deberían poder recargarse en caliente se han movido a la base de datos
• Instrumentación de OpenTelemetry, donde tiene sentido, y un panel de Grafana
• Una API REST incorporada para la gestión de servidores
• Un servidor MCP integrado para LLM
• Nidrah AI, un agente de IA para facilitar la gestión del servidor
• Vista del mapa del servidor en tiempo real
• Auditoría de chat y espionaje en vivo
• Un nuevo motor de jugadores falsos con un sistema lógico Node y un nuevo planificador LLM para cualquier comportamiento
• Sólo mira el vídeo

 

Si hay interés en esto y estoy contento con ello, o me aburro (lo cual siempre hago), lo publicaré en código abierto.

Déjame saber lo que piensas y si hay alguna característica que te gustaría que implementara.

Lo vendes? necesito todo eso, esta super hermoso tu trabajo.

[`NeverMore]

τι εκανες εκει παλι ρε τρελάρα; 

 

welcome back mate, happy seeing you online again, well thats beyond l2 needs for sure and It’s rare to see anyone pushing Interlude this far technically anymore without trying to monetize it.

definitely interested in seeing the source whenever you're ready to share it!

keep it up!