Reused Mobile Devices

Server and Client for running an automatically managed llama RPC cluster.

Run Server

go run .

Web UI

The long-term frontend is now being moved to React/Vite under rmcluster/frontend.

The current server-rendered pages remain available for transition and fallback. The currently available pages are:

• / for the landing page and navigation
• /dashboard for connected device status

For the new frontend, run the React dev server from rmcluster/frontend and point it at the API routes under /api/ui.

Metadata Cache

Hugging Face model metadata is cached in an embedded BoltDB file.

• Default path: $XDG_CACHE_HOME/rmd/metadata.db (Linux), ~/Library/Caches/rmd/metadata.db (macOS)
• Docker default path: /var/cache/rmd/metadata.db
• Override path with RMD_METADATA_DB_PATH

For Docker, mount /var/cache/rmd as a persistent volume so metadata survives container restarts.

Local Model Storage

Uploaded local .gguf models are stored under:

• Default path: $XDG_DATA_HOME/rmd/models (Linux), ~/.local/share/rmd/models (fallback)
• Override path with RMD_MODEL_STORAGE_DIR

Run Client

The client will announce itself to the tracker and run the rpc server. Replace /path/to/rpc-server with the path to the rpc server binary. Replace 127.0.0.1:4917 with the ip and port of the tracker.

go run ./cmd/linux-client/ -cmd /path/to/rpc-server -tracker 127.0.0.1:4917 -- -c