Release 1.5.2


Hello everyone. Today I’m releasing a long awaited patch. It fixes bugs, expand the number of compatible graphical cards and greatly improve the performances.

With all these changes you will have to redownload the AIs! You can delete the old ones.



Here is a little summary of the improvements

  • No more memory peak when the AI is loading.
  • The possibility to specify how much memory you allow (on both CPU and GPU).
  • The option to use a folder to write data instead of storing it in the memory (It’s called offloading).
  • Many new options for those who like to experiment.
  • AI split into parts of 200 MB to allow a better memory management.


  • Added 8 options. Here is a list of the most useful:
    • Offload model (Instead of running out of memory, the server will write on your disk)
    • limit_memory (Allows you to specify a maximum of memory for the server*)
      • gpu_max_memory (The maximum VRAM allowed)
      • cpu_max_memory (The maximum RAM allowed)
  • All the files needed for a model are now saved in the same folder.
  • Added accessible_regular font for better accessibility.
  • Updated the credits.
  • Fixed the server’s logs never being displayed in the client.
  • Fixed an error when loading scenarios from the Steam Workshop.
  • Added the GPT4All-J model (Please, give me feedback about this model :))
  • The limit of memory isn’t perfect. It’s a limit only for loading AIs. During the generation you will go above it.


  • The AIs have all been modified (cut into small parts) to optimize their performances.


  • Updated the dependencies (It makes compatible many graphical cards).
  • The server is now capable to use both the CPU and GPU at the same time. Greatly improving the performances for users owning a compatible GPU.
  • Updated the config file to give more control over the server when in self-hosting mod.

You can find a full changelog of the server on GitHub