Set the rules for every model call

An AI agent control plane is the layer between your agents and the models they call. It applies your rules to every call: how much the organization may spend, whether Opper stores traces and for how long, which providers and models are allowed, which content is flagged, redacted or blocked, and which provider a pooled model tries first. Rules are set for the organization or for a project, so they hold across every app and agent that runs there.

An LLM gateway is the request path: one API in front of many models, with fallbacks and one bill. The control plane is the set of rules that gateway enforces on each call. On Opper both share one API and one key, so you start with the gateway and turn on Rules when you need them.

Five kinds. Spend limits cap monthly spend for the organization, per project, and per role, user or key. Data retention sets whether Opper stores traces and for how long, from 30 days down to zero, and a provider data policy extends zero data retention to the model provider. Model access restricts providers, service routes, inference and storage locations, makers, maker countries, provider countries and individual models, and can require providers that do not train on or log your data. Checks flag, redact or block content during a call and score answer quality after it. Routing sets which provider a pooled model tries first. Organization rules set the default and project rules can only tighten them.

Yes. Set Opper retention to 0 days so Opper stores no prompts, responses or files, and set the provider data policy in Model access to Zero data retention so only providers that neither train on nor log your content are eligible. Both are rules, so they hold for every call in the organization or project they are set for, on every v3 endpoint. Usage metadata for billing, such as cost, tokens and model, is still recorded; content never is.

Because the Opper API already runs in Stockholm, and because "EU" is more than one question. An EU endpoint or a model suffix covers the calls that use it, and one missed call site sends data elsewhere. A model access rule is set once for the organization or a project and holds for every call, whichever SDK, agent or model name sent it. Inference location, storage location and provider country are separate fields, each set to the EU, the EEA or single countries, a model served by several providers only runs on the routes that pass, and any other call is refused with a 403 before it reaches a provider. On any plan you can also pick an EU route per call by its model id, such as aws/claude-opus-5-5.

Rules cover the technical controls that GDPR, Schrems II, the EU AI Act and DORA ask about. GDPR: checks redact personal data before it reaches the model, Opper keeps traces from 30 days down to none, a provider data policy extends zero data retention to the model provider, and one data processing agreement covers every model. Schrems II: a model access rule keeps inference and storage in the EU, so nothing is transferred across the Atlantic unless you allow it. EU AI Act: traces show what each call saw, what was redacted or blocked and which model answered. DORA: failover between providers and spend limits per organization and project, with measured uptime published.

Yes. Rules apply to every LLM and embedding call made through the Opper v3 API, including the OpenAI, Anthropic and Google compatible endpoints and the playground. Spend limits also cover image, speech and video generation, and model access covers every model type, including image generation, speech, transcription, video, OCR, realtime voice and rerank. Checks do not yet apply to those modalities.

Yes. Every call is recorded with the model that served it, the provider, tokens, cost, latency and status, and usage records are kept for five years for billing and reporting. With tracing on, the full trace also holds the inputs, outputs and every rule that fired, for the retention period you set.

Yes. The Management API lists, creates, updates and deletes the same rules the platform shows, using a management key with the controls:write scope. A rule created from CI or infrastructure-as-code is enforced from the next request.

Opper is pay as you go with no subscription and no markup on token rates. Rules are part of the Control Plane tier, priced as a slightly higher fee on credit purchases. LLM checks are billed at a flat rate per check on top. Current rates are on the pricing page.