>Tutorial: CLI Command Reference

Guided onboarding in five steps: [1/5] hardware detect (determines tier), [2/5] Ollama check (waits if starting), [3/5] model recommend and pull (largest fitting 90% of VRAM, or 50% RAM on CPU; any Ollama model works), [4/5] node key generation (separate low-value secp256k1 key, never the staking wallet), [5/5] prints a prefilled Console staking link. After staking, polls the coordinator and auto-detects the nodeId, saves it to config, then offers service install. Idempotent: re-running resumes from where it left off.

Detects CPU cores, memory, and GPU (NVIDIA via nvidia-smi, Apple Silicon via Metal, AMD via ROCm), confirms Ollama is running, and prints the tier the machine maps to. CPU is tier 1; GPU is tier 2/3/4 by VRAM. Used by setup at step 1/5; also available as a standalone command for the manual path.

Generates the separate, low-value secp256k1 node key at ~/.parallelix/node.key (mode 0600), writes config.json, and prints the nodeKeyHash you register on-chain. Used by setup at step 4/5; also available as a standalone command for the manual path. Refuses to overwrite an existing key without --force.

• --wallet 0x.. Record the staking wallet for reference (the CLI never signs with it).
• --force Overwrite an existing node key. Backs up the old key first, but the old node is orphaned.

Lists the models installed in the local Ollama runtime, pulls one, shows the curated catalog, or recommends the best fit for the machine. Installed models are advertised in heartbeats; the ParalleliX AI model picker lists what online nodes serve.

• pull <name> Pull a model into Ollama. First pull downloads it. Bare `models` just lists installed models.
• catalog List the curated model catalog with VRAM requirements (llama3.2 3B/4GB through llama3.3:70b 40GB).
• recommend Recommend the largest model fitting 90% of VRAM (50% RAM on CPU). Any Ollama model works; the recommendation is a starting point.

Runs the daemon in the foreground: attaches to the on-chain node, sends signed liveness heartbeats every 10s (your uptime), polls for dispatched inference requests, runs each whole through Ollama, and returns the result with a SHA-256 Proof-of-Execution. Shows a TTY dashboard (status, models, lifetime earnings, request feed) unless --plain or a pipe is detected. Self-stops cleanly on not_staked or in_cooldown. Tolerates transient 403s for ~5 minutes. Drains on SIGINT / SIGTERM.

• --node-id <id> The on-chain nodeId. Optional if setup already saved it to config.
• --gpu | --cpu Force the execution mode. Defaults to gpu when a GPU is detected.
• --model m Prefer a specific installed model.
• --plain Force line output instead of the TTY dashboard. Used automatically under systemd or pipes.

Installs the node as a managed background service that survives logout and reboot and auto-restarts on failure: systemd on Linux (ordered after ollama.service, Restart=on-failure), launchd on macOS. On Windows, run the node inside WSL2 where the Linux path applies.

• --node-id <id> The on-chain nodeId. Optional if setup already saved it to config.
• --gpu | --cpu Force the execution mode for the installed service.

Checks parallelix.io/cli/manifest.json for a newer version, downloads it, verifies the sha256 checksum, gates on a node --check (confirms Node version is compatible), then performs an atomic swap: old binary backed up as .bak, new binary swapped in. Aborts untouched on any checksum mismatch.

Runs diagnostics with remediation hints: Node version, node key presence and permissions, config validity, detected tier, Ollama running, models installed, and coordinator reachability. Exits non-zero if any check fails.

Prints the local node state from the config: node address, nodeKeyHash, nodeId (if saved), tier, coordinator URL, Ollama URL, and the staking contract address.

Prints the last lines of the node log at ~/.parallelix/node.log: starts, served requests with PoE prefixes, heartbeat status, and exit reasons. Under systemd, use journalctl -u parallelix-node -f for a live feed.

Prints the CLI version. The reference version documented here is 2.0.0.