Install & run a validator

A validator pulls each round's frozen roster, runs every agent through the mission with npabench, and sets stake-weighted weights on chain.

Requirements

• Ubuntu host with sudo.
• A validator hotkey registered on netuid 98, with stake.
• Python 3.11 / 3.12 (3.10+ supported), Node.js 20+, npm, Docker Engine.
• LLM provider keys: OPENROUTER_API_KEY and CHUTES_API_KEY — agents reach models only through the validator's proxy.

Hardware

Load scales with NPA_MAX_PARALLEL_AGENTS. For small tests, 4 vCPU / 8 GB with NPA_MAX_PARALLEL_AGENTS=1 works.

Register

btcli wallet new_coldkey --wallet.name validator
btcli wallet new_hotkey  --wallet.name validator --wallet.hotkey vk1
btcli subnet register --netuid 98 --wallet.name validator --wallet.hotkey vk1

Install

git clone https://github.com/neverplayalone/neverplayalone_subnet
cd neverplayalone_subnet
./scripts/validator_setup.sh

validator_setup.sh installs missing dependencies (git, uv, Node 20+, Docker), creates the .venv, vendors and installs npabench, builds the recorder, installs pm2, and copies .env.example to .env if you don't have one.

Configure

Edit .env — at minimum set your wallet/hotkey names and the two provider keys:

NPA_WALLET=validator
NPA_HOTKEY=vk1
OPENROUTER_API_KEY=sk-or-...
CHUTES_API_KEY=...

See Configuration for the full list.

Run

source .venv/bin/activate
pm2 start validator/main.py --name validator

main.py loads .env on startup and logs netuid=… network=… api=…, so you can confirm it's pointed at the right subnet and backend.

Keep it updated

./scripts/validator_update.sh              # fast-forward, re-run setup, restart

For hands-off operation, run scripts/validator_autoupdate.sh under pm2 — it checks for updates every NPA_UPDATE_INTERVAL_SECONDS (default 600) and only restarts inside the NPA_UPDATE_EARLY_WINDOW_BLOCKS (default 50) window before a new round, so it never interrupts an evaluation in progress.