Before you install the AI model server

  • Confirm your machine meets the system requirements documented in FileMaker Technical Specifications. It should be a dedicated machine reserved for use as an AI model server. Don't use an email, print, or network file server.

  • To prevent performance issues, AI Services should run on a separate server, not your main FileMaker Server.

  • Disable sleep and hibernate modes, hard drive sleep, and screen savers.

  • The AI model server requires a web server in all deployments. The web server hosts the web-based Admin Console application and handles some data transfer tasks. The AI model server requires that port 80 is available for web connections and port 443 is available for secure web connections. If the installer detects existing websites using these ports, you must disable those websites and make the ports available.

    • Windows: The FileMaker Server installer enables the IIS web server if it isn't already enabled, then runs the Microsoft Application Request Routing (ARR) installer, creates its own website in IIS, and configures the website to use the ports specified for web connections.

    • macOS and Linux: The web server does not need to be enabled. If it is enabled, ensure that no existing website uses port 80 or 443. The installer creates a separate web server instance and enables it for the host to use on these ports.

      See Setting up the web server.

  • In the server computer's firewall, open ports 80 and 443 so that the AI model server can communicate with administrators and clients.

  • Locate your license certificate. See About the license certificate.

  • For security reasons, use a custom SSL certificate. Import your own SSL certificate purchased from a trusted certificate authority.

  • If the machine has antivirus software installed, you may need to disable or uninstall it before running the FileMaker Server installer. Don't enable antivirus software again until after the installer has finished.

    Do not allow antivirus software to scan the folders that contain hosted databases or the folders that contain files for container fields that store data externally.

  • You cannot run two different versions of Claris AI Model Server on the same machine at the same time.

  • Because some DHCP servers cycle IP addresses, use a static IP address.

  • Text-embedding operations are generally less demanding than other inference or fine-tuning operations; size your server requirements accordingly.

  • The AI Model Server needs a machine with GPUs:

    • GGUF CPU-only models are not supported.

    • On Windows and Ubuntu: use NVIDIA GPUs and enable CUDA.

    • For macOS, use a Mac with Apple silicon (M-series chips) computers and pick an MLX model or convert a model to MLX.

  • Prioritize having sufficient memory when choosing hardware resources:

    • On Apple hardware: 32 GB of RAM will get you going with the smaller models. 64-96 GB or more give you access to bigger models and more accurate resources.

    • On Windows and Ubuntu: pay attention to the amount of VRAM on the NVIDIA hardware – more is better, and a minimum of 24 GB will work for small to medium models.