Changelog2026-10-02Update

Pools for your own machines

Run Nodus work on machines you already own, before Nodus rents anything.

Enroll a host in one line. nodus create enrollmenttoken --pool lab prints an installer that checks the signed nodusd agent and joins the host to the pool. The token works once and expires after 24 hours. Each host shows up as a Node with its GPUs, utilization and health, and nodus pool utilization lab shows the history.
Route jobs to the pool first. Set placement.pool on a Job. The pool’s routing decides what happens when it is full: wait for it, wait while the market is more expensive, or burst to the market, with an approval if you ask for one. Your machines cost $0; routing costs $0.02 per device-hour of Nodus-scheduled work.
Forecasts and recommendations. Turn on Predict for a pool ($99.00 per pool per month) to see demand and free capacity with p50 and p90 bands, and recommendations for sizing, idle hosts and wait policies.
Routine actions, with approvals. spec.act reclaims idle hosts, drains in maintenance windows and tunes wait policies. Start in shadow mode to see what would run, approve each action, or let policies run on their own. nodus pool pause stops every action at once.
Cloud accounts, read-only. Connect AWS through a CloudFormation role or GCP through Google sign-in to list your GPU instances and spend, and enroll an instance into a pool. Nodus never creates, changes or deletes anything in your cloud account.