Opper.ai
Opper provides one API for accessing models from multiple providers. Its official gateway page positions the service for developers and agent applications, with hosting in Stockholm and a choice of provider routes. The publisher lists more than 700 models; actual availability, price, and data handling depend on the selected model and route.
Connect applications and choose routes
Opper supports OpenAI-compatible requests, including streaming, tool calls, and provider-native prompt caching. Its pricing FAQ describes changing the base URL and API key to migrate an existing compatible integration. Developers can specify a model or provide a fallback list so a failed request is retried with the next model.
The Management API covers projects, runtime keys, members, and dynamic routes. An MCP server also exposes management operations to coding agents. Organizations without a payment card have published management limits of 60 requests per minute and three active management keys.
Add organization controls
Control Plane adds organization budgets, spending caps, model allowlists, routing rules, and input or output checks. The checks can flag, redact, block, or score content. Writing management rules requires Control Plane, even though the underlying Management API is available on every plan.
Retention needs a route-specific reading. Opper says Gateway stores usage metadata rather than prompts or responses. Control Plane defaults to zero retention but can store traces when you enable tracing, with a retention window of one to 30 days. Provider policies remain separate: some zero-retention Azure and Google Vertex routes require a signed Enterprise agreement. EU platform hosting alone does not mean every selectable provider route stays in the EU.
Understand the credit fees
According to the captured pricing FAQ, Gateway charges a 3% fee when you buy credits. Control Plane raises that purchase fee to 5.5%. The publisher says model token rates pass through at provider cost, while LLM checks incur an additional flat charge per check. VAT is added at checkout according to billing country.
Free models work without a payment card. Premium models use credits, with no minimum spend stated. Enterprise customers can request custom terms and deployment options. For budgeting, include model usage, the credit-purchase fee, and any enabled checks rather than treating the platform fee as the complete inference cost.