Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.opper.ai/llms.txt

Use this file to discover all available pages before exploring further.

Control Plane features are in early access and need to be turned on per account. Contact support@opper.ai if you’re interested.
Comply is the layer that constrains how a call can run: which models it can reach, how long its trace is retained, how much you’ll spend, and whether it has to stay inside a Zero Data Retention provider set. You assemble those constraints as rules, each scoped to your organization, specific projects, or specific functions.

​
Rule kinds

Open platform.opper.ai and navigate to Controls → Comply to add a rule. There are four kinds.

​
Model allowlist

Constrain which models calls can reach across four dimensions. Leave a dimension empty to skip it.
  • Allowed providers — anthropic, openai, google, etc.
  • Allowed regions — eu-west-1, us-west-2, etc.
  • Allowed countries — including an “Other” bucket for models without country metadata. Useful for data residency.
  • Allowed models — explicit model names like anthropic/claude-opus-4-6.
A live counter shows “N of M models match.” so you can verify the rule’s reach before saving. When a child rule inherits from a parent (project allowlist inheriting from an org allowlist, for example), items the parent disallows appear dimmed in the picker with a tooltip like “Not allowed at organization scope — broaden the parent allowlist first.” A child rule can only narrow what the parent permits, never widen it.

​
Trace retention

A single number, 0–30 days (default 30). Generations and spans inside the rule’s scope are kept for that many days, then deleted.

​
Budget

  • Monthly limit — a dollar cap.
  • Soft alert at X% — default 80. When monthly spend crosses this, email alerts are sent to your org (if alerts are enabled below).
  • Email alerts — checkbox.
Budget rules conflict gracefully — if an org-level budget is tighter than a project-level one, the org limit wins and the platform surfaces a clear error message.

​
Zero Data Retention (ZDR)

No config. Enabling a ZDR rule locks calls in scope to ZDR-eligible providers (today: evroc, berget) and forces retention to 0 days. The rule shows you how many models in the catalog qualify.

​
Scope

Each rule applies to one of three scopes:
  • Organization — every project and every function.
  • Projects — one or more specific projects.
  • Functions — specific functions inside a project.
The allowlist supports all three scopes. Retention, Budget, and ZDR are limited to organization or project scope (no function-level).

​
Scope chain and intersection

Rules at different scopes layer rather than override. The effective allowlist for a function is the intersection of every allowlist that applies to it — org rules narrow what projects can do, project rules narrow what functions can do, function rules narrow further. You can never widen at a lower scope.

​
In traces

Comply events appear on a span with a Scale icon labeled Comply:
  • Passed — the call satisfied every Comply rule in its scope chain.
  • Blocked — a rule rejected the call. The span carries the error; the Comply event identifies which rule.
Each event carries the rule’s name (when set) and a scope badge (Org-level / Project-level / Function-level). When a model is denied by the allowlist, you can see which rule and scope produced the block without leaving the trace view.

​
In playground

Comply rules apply to playground calls when the Project controls toggle in the task sidebar is on. If a rule would block the call, the playground run fails the same way a production call would — the trace shows the Comply event with the blocking rule. To verify a function would still work if a rule were relaxed, switch Project controls off for the run.
Set the strictest policy at organization scope (e.g. “EU-only providers”). Use project and function scopes for narrower exceptions or further tightening. Trying to “loosen” at a lower scope won’t work — the parent constraint always wins.
Pair a ZDR rule with retention=0 on a specific function when handling regulated data. Keep ZDR scoped tightly — it limits the available model set.