Track and cap your AI spend

Know exactly what your AI is costing — by user, model, agent, and conversation — and set hard limits so a run can never overspend.

What you'll use

Cost Analysis in AI Studio for visibility, and cost limits on orchestrations for control. Every message and coding-agent session is metered and priced.

Steps

1. Open AI Studio → Operate → Cost Analysis (a tenant-admin view). See spend broken down by user, by model, by conversation, and by source (in-app messages vs SiloLink coding sessions).
2. Drill into a person or team to see who's driving cost, and check the cache-read rate — a high rate means prompt caching is saving you money.
3. Note the real vs imputed split: metered API calls show real cost; work done on a subscription coding tool shows as imputed (no marginal charge), so you see true economics.
4. To cap spend, set a cost limit when you launch an orchestration — the run stops itself if agent spend hits the limit. Give scheduled agents a daily cost limit for the same protection.

Result

No surprises. You can attribute spend to the work and the people that caused it, prove caching is paying off, and guarantee a runaway pipeline can't blow the budget — it halts at the limit and waits for you.

Cost attribution rolls up by who initiated the work, so a coordinator's launched pipeline is credited to them — handy for chargebacks across teams.