Usage and billing¶
Silico measures organization usage in Research Units (RUs). Billing should help a team make research tradeoffs, not turn every experiment into bookkeeping.
Open Organization settings → Billing to understand the balance, see where usage came from, and set safeguards. Then use plans and experiment budgets to make the cost of individual research choices visible before they scale.
Available actions depend on billing configuration and role.
Understand the balance¶
RU Balance shows:
- Remaining RUs
- Consumed RUs relative to the total credited balance
- the approximate dollar value represented by usage
- the percentage of total RUs consumed
RUs represent charged Silico usage, not GPU-hours. Thread conversation, planning, agent work, and experiment execution can all contribute. An experiment's Compute view separately shows its jobs and runtime, so the two views are not expected to show the same totals. Use Billing to answer “What did the organization consume?” and Compute to answer “What ran, where, and for how long?”
Review consumption¶
Daily RU consumption covers 7 or 30 days. Choose See details for thread and experiment attribution.
Use this view to understand patterns and find expensive work rather than to reconcile every RU individually. A jump may point to a larger sample, stronger model, higher parallelism, repeated retries, or simply more active research. Some activity may be unattributed or omitted from long lists.
Buy RUs and view payments¶
Buy Research Units starts a one-off purchase. View payments opens invoices and payment controls. Purchases can take a few seconds to appear.
Use Contact sales for contract capacity or non-self-serve configuration.
Auto reload¶
Auto reload buys a configured number of RUs when the balance falls below a floor. Set:
- the RU amount to buy
- the balance threshold that triggers it
- a monthly dollar limit
Auto reload pauses at the monthly limit and resets on the first day of the month. The limit excludes tax.
Member spend limits¶
Member spend limits cap each member's RUs. Set a default, add overrides, and export the table as CSV.
Usage measurement starts when the limit is set. Treat it as a safeguard, not lifetime accounting.
Per-experiment budget¶
An experiment RU budget pauses Silico at the threshold and asks whether to continue. It does not stop existing cluster jobs; cancel those in Compute.
Think of the budget as a decision point rather than a hard downstream compute cutoff. Set it where you would genuinely want to reconsider the evidence gained so far, not at an arbitrary round number no one plans to review.
When the balance runs out¶
When credit enforcement is enabled, zero RUs pauses new billable work. Existing threads and reports remain readable.
Keep usage predictable¶
Before expensive work, review sample size, model, accelerator, parallelism, runtime, stopping conditions, and budget. Ask what result from a pilot would justify the full run. A small pilot is most valuable when it has an explicit scale-up decision, not merely because it is cheaper.
Use Billing to monitor organization-wide RU consumption and the experiment's Compute view to monitor the underlying jobs.