Settings

Integrations and operations

Manage model access, compute, and operating limits. Portfolio governance remains separate from developer execution and production release authority.

At a glance

The controls on this page change model policy or hourly compute state. Project and run evidence remain in the Studio so each operating action stays separate from source editing and deployment.

COST

Operate compute →

DATA BOUNDARY

Audit provider policy →

CHANGE

Operations do not approve source

Compute and provider switches are distinct from candidate approval. Inspect governed run evidence before recording a review head or deliberately reconciling it into canonical source.

Open evidence room →

Open project directory → See next-turn routing → Audit the authority model →

Permission is separate from authentication

These switches admit only adapters with a reviewed disposable execution target. Unconnected experimental adapters stay visible for evaluation but cannot be enabled or selected.

Control-plane posture

API

Observed compute

Retained volume

Cost posture

Return to projects → See the work that may need compute → Review billable deployment layers →

Operating guardrails

The API accepts only tokens issued for this application client and then requires ProjectOwner or PlatformAdmin.
Administrative endpoints additionally require PlatformAdmin.
GPU jobs remain impossible until the optional Batch stack is connected and the account’s GPU Spot quota is raised from zero.
Every CPU or GPU worker has a one-hour hard timeout and one attempt. Authenticated private sessions normally stop sooner: on explicit close, sign-out, or a received page-exit signal, and after 15 minutes without a completed run.
There is no NAT gateway, load balancer, OpenSearch domain, RDS instance, or permanently running EC2 instance.

Trace each guardrail to a trust boundary → Review AI change governance → Inspect the run record behind a change →

The account’s 31 July 2026 Spot sample in us-east-1 ranged from about $0.29/hour for g4dn.xlarge to $0.96/hour for g5.xlarge, before storage and public IPv4 charges. Spot prices vary by Availability Zone and time; verify immediately before enabling a run.

aws ec2 describe-spot-price-history \
  --profile categorise --region us-east-1 \
  --instance-types g4dn.xlarge g5.xlarge g6.xlarge \
  --product-descriptions 'Linux/UNIX' --max-items 20