Gaianet Node Executor

Problem

Gaianet’s decentralized AI inference network requires operators to run edge nodes that serve LLM inference requests and register Node ID / Device ID against the network’s web dashboard. The stock install flow mixes package installation, gaianet init, and a long-running foreground process across both Ubuntu/Debian and macOS — error-prone to repeat and difficult to keep alive across SSH disconnects.

Approach

• OS-branched scripts instead of a single conditional blob: dedicated entry points for Linux (apt) and macOS (brew) to keep package-manager logic isolated.
• screen-based supervision so gaianet start survives terminal disconnects and remains attachable for log inspection.
• Official installer passthrough: wrap Gaianet’s install.sh rather than re-implementing the bootstrap, minimizing drift against upstream releases.
• Explicit init step (gaianet init) before start, so model/config download failures surface before the screen session detaches.

Implementation

Bash scripts targeting Ubuntu/Debian and macOS. The Linux path updates apt, installs prerequisites, runs Gaianet’s official install.sh, executes gaianet init, then opens a screen session running gaianet start with auto-attach for live log tailing. The macOS path mirrors this via Homebrew. After startup, the operator copies the printed Node ID / Device ID into the Gaianet web dashboard to register the node against their account.

Outcome

• One-command bring-up across both supported OSes with the long-running process correctly detached under screen.
• Real-time log visibility preserved via screen attach, enabling post-mortem on inference failures without re-running setup.
• Node registered to the Gaianet dashboard and serving AI inference requests on the network.

Technologies

• Scripting: Shell (Bash)
• Infrastructure: Linux (Ubuntu/Debian), macOS
• Service: Gaianet AI Node
• Management: screen session