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.

Getting a single AI call to work is easy. Making AI reliable in production is not. Teams that ship AI into real products quickly run into the same set of problems: you’re locked into one provider’s SDK, switching models means rewriting integration code, there’s no visibility into what the AI is actually doing, and quality is unpredictable. As you add more models and more use cases, the complexity compounds — separate credentials, different APIs, no shared observability, and guardrails bolted on as an afterthought. Opper exists to solve this. It sits between your application and the models it uses, giving you a single place to connect, control, and observe all your AI interactions.

​
Gateway — multi-model, multimodal generation

The gateway is your single connection to AI models. One API, one key, 200+ models across all major providers — OpenAI, Anthropic, Google, Mistral, and more.
  • Any model, same API. Route to any supported model without provider-specific SDKs or credentials. Switch models with a config change, not a code rewrite.
  • Multimodal by default. Text, images, audio, embeddings — the same API handles all modalities. Build applications that work across media types without juggling separate integrations.
  • Model independence. Models change fast. New ones launch, pricing shifts, capabilities improve. Your code shouldn’t have to change with them. With Opper, the model is a parameter — everything else stays the same.
  • Drop-in compatibility. Use the Opper Multimodal API, or connect through OpenAI, Anthropic, and Gemini compatible endpoints. Migrate incrementally without rewriting your stack.

​
Control Plane — making AI reliable

Control Plane features are in early access and need to be turned on per account. Contact support@opper.ai if you’re interested.
The control plane is where you go from “it works” to “it works reliably.” It handles observability, routing, guardrails, compliance, quality steering, and memory — the things that matter once AI is in production.
  • Observe. Score every generation against a configurable judge. Pick the rigor (Fast / Balanced / Thorough), the sample strategy, and either a 0–1 score or a binary verdict. Results land on the span next to latency and cost.
  • Route. Pin a default model for a scope. Calls that omit model use it; calls that pass an explicit model are still subject to the Comply allowlist.
  • Guard. Run guardrails at the gateway, before data reaches the model and before responses return. LLM guards use a custom prompt to classify content; Regex guards match known patterns. Each guard can Flag, Block, or Redact.
  • Comply. Constrain which models calls can reach (provider / region / country / model allowlist with parent-intersection across scopes), trace retention, monthly budget with email alerts, and Zero Data Retention.
  • Steer. Coming soon — Steer will use Observe scores to select few-shot examples and optimize prompts. The SDK feedback flow is available today and surfaces on the span.
  • Memory. Store and retrieve knowledge through semantic indexes. Give your AI access to custom context without managing vector databases yourself.
Each Control Plane feature is a list of rules. A rule has a config, an optional name, and a scope — organization, specific projects, or specific functions. You set rules in Controls → <Feature> in the platform, see them fire at runtime in the trace view (each control event is labeled with the feature, the rule name, and a scope badge), and test how they affect a call in the playground — a Project Controls toggle in the task sidebar applies or skips your project’s rules for the run.

​
Security

Opper is a secure intermediary between your application and external models. All traffic is encrypted and authenticated, and the platform is designed so that your data stays under your control.
  • Infrastructure. Opper deploys regionally on AWS. All data except model calls is contained within the deployment — vector databases, routing, tracing, and caching stay in-region. Currently available in Stockholm, Sweden.
  • Authentication. SSO is supported with major providers (Google, GitHub). All API calls are authenticated with API keys. Users get their own account and can be part of a multi-user organization.
  • Compliance. Opper follows European data protection directives and security standards, including GDPR. For compliance details, contact support@opper.ai.