Skip to navigation

Usage

See what an account has spent and created this period, and request more of a capped resource
View as Markdown

Open Usage in your account settings (or an organization’s settings, for an org account) to see what the account has spent and created this billing period.

Spend this period

The page totals spend against the account’s spend limit, split into two streams:

  • Compute — agent runtime across every deployed agent, plus the sandboxes agents run code in.
  • Models — LLM usage through the AI Gateway.

Below the total, a chart breaks spend down by day, in US dollars, across both streams.

How compute is priced

Compute is billed per CU-hour. A CU (compute unit) is max(vCPU, memory in GB ÷ 2), multiplied by replica count.

The deploy form estimates a monthly cost as you size the agent and each component it declares. The estimate assumes one replica and excludes storage; scaling to more replicas multiplies your actual charge accordingly. When the agent is your account’s free one, the form shows $0 and the price it would otherwise cost.

Scheduled and one-time ingestion runs (cron and startup jobs) bill per run, for the wall-clock time each run reserves its resources, including a run that fails or retries.

The messaging sidecar (web chat, Slack, Teams) is not billed. Every messaging-capable agent reserves one by default, and its compute is free.

How sandboxes are priced

A sandbox is billed per CU-hour for the time it runs, at its own sandbox rate. Its CU follows its memory, at 1 CU per 2 GB:

SandboxMemoryCU
Default2 GB1
With browser4 GB2
With browser and size: large8 GB4

Billing starts when the sandbox launches, so the time it spends installing your declared software counts. It stops when the sandbox is suspended or removed, and a suspended sandbox costs nothing. A sandbox your local agent uses under ast project start is billed the same way as a deployed agent’s.

The free agent allowance below does not cover sandboxes. A sandbox that belongs to your free agent is billed like any other.

Your first agent’s compute is free

Each account’s oldest running agent is free, whatever size you give it. Its CU-hours still appear in your usage breakdown, rated at $0.

The free agent is the running agent you created first. Delete or stop it, and the allowance passes to your next-oldest running agent. The allowance covers compute only, not model usage through the AI Gateway.

Resource limits

Resource limits lists what the account has created against what it’s allowed: agent builds, deployments, blueprints, members, knowledge stores, and PrivateLink endpoints. A resource with no cap shows as unlimited.

Owners and admins can request a higher limit for one of these resources. A request starts pending; the account’s past requests and their status (pending, approved, or denied) stay listed in the same section.

Who can see this page

Every personal account’s owner can see their own usage. For an organization, only admins and owners can open Usage or request a limit increase; other members don’t see it in navigation, and a direct link shows an access-denied message instead of the page.

Next steps

⌂ 换网址