Free while it is a preview.
There is no billing yet, and inventing three tiers before anyone has run this in anger would be theatre. Here is what a harness actually costs to run, so you can work out what it would cost you.
What it costs to run
Billed by the second, when it exists.
A harness is one short-lived task. There is no pool of agents waiting for work, so there is nothing to pay for between alerts.
small
$0.016
per hour · 0.5 vCPU · 1 GiB
medium
$0.032
per hour · 1 vCPU · 2 GiB
A two-hour triage session on a small harness is about three cents of compute. Model tokens are billed by your provider, on your own key. We never resell them.
Bounded by design
Nothing runs longer than you allowed.
0
inbound network rules to a harness
Not restricted. Absent.
4h
default time to live
Server caps it at 8h. Not optional.
1
session summary that can never be edited
Written at teardown. Not by you, and not by an administrator.
Included
Everything, while it is a preview.
- Unlimited templates and automations
- Up to three concurrent harnesses per workspace
- Every model provider: bring your own key or sign in with a subscription
- Roles, teams, scoped permissions and audit
- Browser terminal and workspace editor
When pricing does land it will be usage-based and legible: harness-hours and automation runs, attributed per team, with the same numbers you can already see. Anyone using it during the preview will get notice long before anything changes.