A workload owns the settings for a team, user, agent or application: model access, limits and budgets. API keys are credentials underneath it.
Workloads: organise application access and its credentials.English interface · illustrative demo data. Open the image for full size.
Set the allowed models, requests and tokens per minute, and a daily, monthly or total budget. Then choose what happens when the budget runs out:
block stops over-budget calls with 402.
notify lets them continue.
Answer timeout
Under Limits & budget, the answer timeout sets how long a model call waits for the provider's answer: 10 to 600 seconds, 300 by default.
A call without an answer in time fails with 504 and is not replayed, because the provider may still be working on it.
A caller can shorten the timeout per request with the x-sluis-upstream-timeout header, never extend it.
A streamed answer that has started is not cut by the timeout. Only 600 seconds of silence mid-stream end it.
Workload limits and budget, including the provider answer timeout.English interface · illustrative demo data. Open the image for full size.
Overrides, pausing and keys
An empty option_overrides inherits the organisation defaults.
Choose an available AI recognition model and override individual entity kinds and built-in detectors for this workload, including document and image inspection.
Pausing a workload refuses its requests (403) without revoking credentials.
You can rotate keys without changing the workload's policy.
Workload policy overrides: choose deviations from the organisation defaults.English interface · illustrative demo data. Open the image for full size.
Scope guard
The scope guard checks every model call made with the workload's keys (embeddings excepted) against a scope you write in plain language: what this workload's AI may and may not do, up to 4,000 characters. Choose a mode:
off skips the check.
log judges in the background without delaying the call, and records out-of-scope calls in the audit log.
block waits for the verdict and refuses an out-of-scope call with 403 scope_violation before it reaches the provider. If the check cannot complete, the call is refused with 503 scope_guard_unavailable.
How it judges
A small EU-hosted model (Mistral Small) judges each call after data protection, on the text the provider would receive.
Each check is a separate audit receipt, billed at the model's token price.
Your residency policy must allow that model, or the guard cannot be switched on.
Input made only of files or media the guard cannot inspect is refused with 503 in block mode, and skipped in log mode.
Workload scope guard: Log mode and the saved scope checked for each call.English interface · illustrative demo data. Open the image for full size.Console