Getting Started#

Install#

The quickest way to get started is the install script:

curl -sSL https://www.nixi.sh/install | bash

This will:

Detect your system architecture, GPU, and available memory (including APU shared memory)
Download the latest pre-built binary for your platform
Ask if you have an existing LLM server or want to install Ollama locally
Select the best model for your hardware (uses total RAM for APU systems)
Configure GPU acceleration (including HSA_OVERRIDE_GFX_VERSION for AMD APUs)
Write a config file and verify everything works

You can also pin a specific version:

NIXI_VERSION=v0.1.0 curl -sSL https://www.nixi.sh/install | bash

Supported Platforms#

Pre-built binaries are available for Linux on the following architectures:

Architecture	Binary	Common hardware
x86_64 (amd64)	`nixi-linux-amd64`	Standard PCs, servers, most cloud VMs
x86 (i386)	`nixi-linux-386`	Older 32-bit systems
ARM64 (aarch64)	`nixi-linux-arm64`	Raspberry Pi 4/5, Apple Silicon VMs, ARM servers
RISC-V 64	`nixi-linux-riscv64`	RISC-V boards (VisionFive 2, Milk-V, etc.)

The install script auto-detects your architecture. Binaries are statically linked with no external dependencies.

NixOS Flake#

Add Nixi to your flake inputs:

{
  inputs = {
    nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
    nixi.url = "git+https://codeberg.org/ewrogers/nixi";
  };

  outputs = { nixpkgs, nixi, ... }: {
    nixosConfigurations.myhost = nixpkgs.lib.nixosSystem {
      modules = [
        nixi.nixosModules.default
        {
          services.nixi = {
            enable = true;
            package = nixi.packages.x86_64-linux.default;
            llmUrl = "http://localhost:11434";
            model = "qwen3:30b-a3b";
          };
        }
      ];
    };
  };
}

Build from Source#

git clone https://codeberg.org/ewrogers/nixi.git
cd nixi
go build -o nixi ./cmd/nixi

Connect to an LLM#

Nixi supports any OpenAI-compatible API. It auto-detects the protocol from the URL.

Ollama (local)#

nixi --url http://localhost:11434 --model qwen3:30b-a3b

LM Studio#

nixi --url http://localhost:1234 --model qwen/qwen3-30b-a3b

Any OpenAI-compatible server#

nixi --url http://your-server:8080 --model your-model --api openai

First Run#

TUI (Terminal)#

Launch the TUI in your terminal:

nixi

You will see:

A header bar with your hostname, architecture, and active model
A text input at the bottom
Type a message and press Enter to chat with Nixi

Web UI (Browser)#

If you enabled the web UI during installation, it’s already running. Open http://your-host:6494 in your browser and log in with the password shown during setup.

The web UI has full feature parity with the TUI – streaming responses, slash commands, autocomplete, confirmation prompts, and the same icy theme.

To start it manually (if you skipped it during install):

nixi serve

On first run, a random password is generated and printed to the terminal. You can also set your own:

nixi auth --set    # prompts for password (empty for random)

Tip: Want a friendly URL like http://nixi.home.local? Just ask Nixi to set up a reverse proxy for you. She can configure Caddy or Nginx and point your local DNS automatically.

See the Web UI guide for NixOS module options, systemd services, and firewall configuration.

Slash Commands#

Type / to see available commands (works in both TUI and Web):

/help – list commands
/model <name> – switch models
/system – show system info
/theme – toggle dark/light theme
/clear – clear conversation
/memories – list saved memories
/nodes – list configured nodes
/adopt <name> <host> – adopt a remote node
/quit – exit (TUI only)

See Tools for the full list of tools Nixi can use to inspect and modify your system.