Pricing

Most of the hub is free. What costs money is compute we pay for on your behalf: cloud GPUs and the language model behind the Coach, the planner and the agent.

Free

• Hosting public and private repositories: datasets, models, robots.
• Browsing, the episode viewer, the robot Sim tab, the Coach's corpus.
• Everything that runs on your machine: training with lucen worker run, sim rollouts with --local or lucen rollout worker, lucen serve, the MCP server and the robot driver.

Metered

• Cloud GPUs (executor: modal, where this deployment has the backend): billed by the second from when a job claims its machine, at the tier's hourly rate. Each rate is the GPU provider's own price × 1.28; the multiplier covers the CPU, memory and storage around the GPU. The current rates are below. A run's budget_usd is enforced by the hub: it stops the run when the next report would cross it.
• The model (the Coach's diagnosis and plan, the planner, the agent): metered at the model provider's list price. The agent draws on a $5 monthly allowance per account (an organisation shares one); the Coach has a daily request quota per key.
• Hosted CPU work (robot checks, ppo_mujoco training, CPU sim rollouts, where this deployment has the backend): every account gets a $2 monthly allowance, metered at the CPU host's rate; beyond it, run the same thing on your own machine. Cloud GPUs are enabled per account by the operator.

Payment

There is no payment processor yet. Nothing is charged to a card: metered usage is recorded in a ledger at the rates on this page so you can see what it costs, and paid backends are enabled per account by the operator. Before anything is ever charged, this page and the terms will say how, and you will be told in advance.

Prices may change; the rates the API publishes (GET /v1/hardware_tiers) are the ones that apply.