Install the server
The server is Local Cognitive without a window: the same chats, models, tasks and workflows, running as a service on a Linux machine. You use it from the app on your Mac or PC, from anywhere.
Requirements
Section titled “Requirements”- Linux x64 with systemd: Ubuntu 22.04+, Debian 12+, Fedora 36+ or Amazon Linux 2023. That means glibc 2.34+, libstdc++ from GCC 12+ and OpenSSL 3.
- Network: outbound HTTPS only. You don’t open ports or need a domain, a static address or a certificate; the server works behind NAT.
- NVIDIA GPU, optional: driver 570 or newer. Without one, models run on the CPU.
- Disk: about 1.5 GB for the server with the CUDA runtime, plus your models.
Install
Section titled “Install”curl -fsSL https://github.com/Ilyaberdar/local-cognitive-AI-system/releases/latest/download/install.sh | sudo bashThe installer:
- Downloads the newest release and checks its Ed25519 signature and checksums.
- Installs it into
/opt/local-cognitive/releases/<version>with its own Node runtime, and points/opt/local-cognitive/currentat it. - Creates the service user
local-cognitive, the data folder/srv/local-cognitiveand the key that encrypts the server’s secrets. - If it finds an NVIDIA GPU with driver 570+, it shows NVIDIA’s license and installs the CUDA runtime (see NVIDIA GPUs).
- Starts the
local-cognitivesystemd service. - Prints a one-time key to connect your computer.
Options
Section titled “Options”Pass them after bash -s --:
curl -fsSL https://github.com/Ilyaberdar/local-cognitive-AI-system/releases/latest/download/install.sh | sudo bash -s -- --cpu| Option | What it does |
|---|---|
--cpu |
Skip the CUDA runtime; models run on the CPU. Add it later with sudo local-cognitive-server cuda. |
--data-dir <dir> |
Keep the server’s data somewhere else than /srv/local-cognitive. |
--no-pair |
Don’t print a connection key at the end. |
--yes |
Answer every question with yes, including NVIDIA’s license. |
What runs where
Section titled “What runs where”| Path | What it is |
|---|---|
/opt/local-cognitive/current |
The running release; previous is kept for a rollback |
/usr/local/bin/local-cognitive-server |
The server command |
/etc/systemd/system/local-cognitive.service |
The service, run as local-cognitive with a locked-down sandbox |
/srv/local-cognitive |
Chats, models, settings and the work in progress (mode 700) |
/etc/local-cognitive/vault.key |
The key for the server’s encrypted secrets, outside the data folder |
Connect your computer with the key the installer printed.