Opper logo
Api Tools · Llm Providers

Opper

Opper is a European AI gateway routing one OpenAI-compatible endpoint to 700+ models from 30+ providers. Hosted on AWS Stockholm with EU data residency, it suits developer teams needing GDPR-ready inference, spend caps and audit trails.

Active GDPR compliant Pay As You Go API available 16+ Verified by Guidaio
Overview

What is Opper?

Opper is a European AI gateway and LLM router: a single endpoint reaching more than 700 models from over 30 providers, across text, image, voice and video. It is designed as a drop-in replacement for the OpenAI SDK - point the client at https://api.opper.ai/v3/compat, set OPPER_API_KEY, and pass the model as a provider/model string such as anthropic/claude-sonnet-4-5 or openai/gpt-5-mini. The Anthropic, Gemini and Responses clients work against the same base URL. An OpenAPI specification is published and the API reference sits on docs.opper.ai.

Around that gateway, Opper adds an agent layer and a control plane. Python and TypeScript SDKs cover tools, MCP, span-level tracing and multi-agent composition; a CLI (npx @opperai/cli) and an installable skill for coding agents (npx skills add opper-ai/opper-skills) shorten setup. The control plane is organised into named bricks: Observe scores generations automatically with a judge model, Route sets a default model per organisation, project or function, Steer selects few-shot examples and optimises prompts, Guard masks personal data and filters content, and Comply handles allowlists, spend caps and retention controls. Semantic indexes provide storage for retrieval-augmented generation.

Commercially, three offers are published: Gateway at a 3% platform fee, Control Plane at 5.5%, and Enterprise on custom terms. There is no markup on tokens, provider rates being passed through at cost. Automatic failover between providers is offered, and effective uptime is measured and published route by route.

Opper states on its homepage that it serves 50,000+ developers and companies reaching 10M+ users; these are the publisher's own claims, not independent measurements. Customers named on the site include Instabridge (credited with 200M+ downloads across 190+ countries), Svenska Bostäder (28,000+ apartments) and Njord, with further case studies for Ping Payments, Beatly, Get-Tested and AIBob. The same publisher also runs satellite products - AI Roundtable, Media Studio and an LLM Leaderboard - plus /models and /makers directories, none of which are the gateway itself.

What it does

  • Route an LLM call to 700+ models from 30+ providers with a single API key
  • Switch model without changing code, by editing a provider/model string
  • Declare a fallback list per request, and pay only for the call that succeeds
  • Restrict inference to EU-hosted routes
  • Cap the budget per API key, per project and per organisation
  • Mask personal data and filter content on the way in and on the way out
  • Trace every call at span level, with its cost and latency
Audience

When to use Opper / When not to

A quick filter to help you decide if Opper is the right fit.

When to use Opper

  • European product and engineering teams under GDPR that must be able to state, and prove, where inference physically runs
  • Developers already building on the OpenAI SDK, for whom migration means changing a base URL and an API key
  • Teams that want to switch freely between models without renegotiating a DPA, since Opper acts as a single processor for every route
  • Agent builders who need Python and TypeScript SDKs, MCP tools, span-level tracing and multi-agent composition
  • Platform owners and procurement teams that must cap and attribute AI spend per project, team and key, and enforce model allowlists, zero-retention routes and an audit log

When not to use Opper

  • Teams calling exactly one model from one provider, with no residency requirement, no spend controls and no need for an audit trail: Opper itself points them to that provider's own SDK as the shorter path
  • Non-technical users looking for a ready-made assistant, since this is developer infrastructure and you have to write code
  • Anyone who needs a clearly established permanent free plan: the pricing page presents none and the documentation asks you to add credits before your first call
  • Mobile-first or browser-based users, as there is no mobile application and no browser extension
  • Buyers whose security review requires a published SOC 2 or ISO 27001 certificate, which does not appear anywhere on the site
Get started

How to use Opper

A typical end-to-end flow, from setup to results.

  1. Create an account on the Opper platform, for yourself or for a company, then invite team members
  2. Sign in through Auth0 at platform.opper.ai/auth/signin
  3. Top up your credit balance, or bring your own provider keys (BYOK)
  4. Add credits under Billing before your first call: the documentation states that calls are debited from the account balance
  5. Create an API key in platform.opper.ai/settings/api-keys, keeping in mind that a key belongs to a project
  6. Read the key from the OPPER_API_KEY environment variable and never hard-code it
  7. Point your existing OpenAI SDK at the base URL https://api.opper.ai/v3/compat
  8. Pass the model as a provider/model string, for example anthropic/claude-sonnet-4-5 or openai/gpt-5-mini
  9. Or take the assisted route: paste https://skills.opper.ai into a coding agent such as Claude Code, Cursor or Codex and let it write the configuration
  10. Set up payment through Stripe, with manual or automatic top-ups; Enterprise accounts are billed on invoice
Quick read

Pros & Cons

Pros

  • EU data residency is self-serve from the very first call, without going through an Enterprise contract
  • A single processor for 700+ models: no DPA amendment each time a new model is adopted
  • No markup on tokens - provider rates are passed through at cost, and only a platform fee applies on credit purchases
  • Legal and operational transparency: terms of service, DPA, sub-processor list and security overview are published and dated, and effective uptime is published route by route rather than merely described
  • The sub-processor list states, provider by provider, the retention window and whether that provider trains on content
  • Migration from OpenAI or Anthropic is trivial: a base URL and an API key
  • Failed requests are not billed, only the successful call is debited, and self-serve carries no per-seat fee and no minimum commitment

Cons

  • No security certification is published: neither SOC 2 nor ISO 27001 appears on the site, so assurance rests on a self-declared security overview
  • No about page and no team page: neither founders, nor headcount, nor company history are given on the site
  • The existence of a permanent free plan is not established: the terms of service name a Free Tier, but the pricing page presents none and the documentation asks you to add credits before your first call
  • The terms of service also name a Utility Plan that exists nowhere on the site, so the contractual naming is out of step with the commercial offer
  • The platform fee is a percentage, so there is no entry price in euros or kronor to compare, and the terms specify that fees are payable in SEK unless otherwise agreed
  • Advanced features - full tracing, evaluation, guardrails - raise the fee from 3% to 5.5%, and Observe is billed on top on usage
  • The demo is booked through a Typeform modal with no URL of its own, so there is nothing to bookmark or share
Pricing

Pricing & Plans

Opper is sold pay-as-you-go on prepaid credits, and its own charge is a platform fee expressed as a percentage rather than as a monetary amount. Gateway carries a 3% platform fee taken when credits are purchased; Control Plane raises that fee to 5.5% and bills Observe separately on usage; Enterprise is priced on a custom basis, negotiated on volume, with an annual commitment and invoice billing. There is no markup on tokens: model providers' rates are passed through at cost, which means the final cost of an account depends primarily on the providers and models chosen rather than on Opper's fee. Consequently, no entry price is published on the site in any currency, and no lowest price point can be quoted here. Failed requests, and requests routed to a fallback, are not billed: only the successful call is debited. Self-serve carries no minimum spend, no term commitment and no per-seat fee; bringing your own provider keys is free on Gateway and carries a fee on Control Plane. Prices are quoted excluding VAT, which is added at payment according to the billing country, and the terms of service (art. 7.1.2) state that fees are payable in SEK unless otherwise agreed. Payment is handled by Stripe on self-serve and by invoice on Enterprise. Whether a permanent free plan exists could not be established: the terms of service name a Free Tier, but the pricing page presents no free tier and the documentation asks users to add credits before their first call.

Gateway - 3% platform fee taken on credit purchases
  • 700+ models
  • manual per-request fallback
  • provider-native prompt caching
  • metadata-only tracing
  • free BYOK
  • high rate limits
  • credit balance
  • Stripe payment and community support on Discord
Enterprise - custom pricing
  • 700+ models
  • full tracing
  • custom BYOK
  • dedicated limits
  • invoice billing
  • zero data retention available
  • custom hosting regions
  • SSO/SAML
Prices and plans listed above may evolve. Always check the official pricing page before subscribing.
Trust & Privacy

Data, GDPR & hosting

A consolidated view of how Opper handles your data.

GDPR overview

GDPR implementation is concrete and documented. The privacy policy is dated 2026-03-14 and written from an EEA standpoint; a data processing agreement of the same date is published, alongside a sub-processor list dated 2026-08-27. Standard contractual clauses are available on request and frame transfers outside the EEA. The DPA commits Opper to notifying the Customer of a breach within 48 hours (art. 3.2) and to giving 30 days' notice before adding or replacing a sub-processor, with a right of objection (art. 4.2). Data subject rights - access, rectification, erasure, portability - are exercised through support@opper.ai, with a 30-day response window; the supervisory authority named is the Swedish IMY (Integritetsskyddsmyndigheten). Customers may audit once a year, with ten working days' notice and at their own cost (art. 7). Established in Sweden, Opper needs no Article 27 representative.

Who owns the data?

The terms of service (art. 8.2) state that, as between the parties, the Customer exclusively owns and reserves all right, title and interest in its Confidential Information and Customer Data: Opper acquires no ownership of the content you send. Art. 8.1 reserves to Opper the Platform, the Services and Service Usage Data - volumes, frequencies, failure rates and performance, in anonymised and aggregated form. The data processing agreement (art. 1.1) fixes the roles: the Customer is the data controller and Opper the data processor. At the end of the contract, art. 8.2 of that agreement gives the Customer the choice between return or deletion of the personal data held.

Reuse rights

Customer Data remains the Customer's, and Opper states that it does not use customer data to train models - a statement repeated on the homepage, the pricing FAQ, the security overview and the compliance page. What is kept depends on the tier. On Gateway, the content of requests and responses is not stored at all: only metadata (model, tokens, latency, status, cost) feeds analytics and billing. On Control Plane, tracing writes inputs and outputs into an encrypted Amazon RDS database, with retention configurable per project. Enterprise allows configurable retention, including zero data retention. Documents used for retrieval are vectorised into RDS, while the source files sit in a private S3 bucket segregated per organisation. The distinction that matters is between Opper and the model providers behind it: depending on the route chosen, a provider may retain content for abuse monitoring, and Opper publishes those retention and training policies model by model in its /models directory and its sub-processor list. That list names one explicit exception on the provider side: DeepSeek's first-party API, which trains on the content it receives.

Data retention & training

Retention summary
Retention depends on the tier and on who holds the data. On Gateway, neither prompts nor responses are kept: only metadata such as model, tokens, latency, status and cost. On Control Plane, traces including prompts and responses are stored when tracing or Observe is enabled, with retention configurable per project, and zero data retention is available on request. Enterprise allows configurable retention and contractual ZDR. On the model provider side, retention is published per model and per route in the /models directory, ranging from zero to a fixed abuse-monitoring window; the sub-processor list cites AWS holding up to 30 days for abuse monitoring before deletion. Deleting an index deletes its files and index; deleting a function deletes its events. Backups keep daily snapshots five weeks and weekly snapshots fourteen months. At contract end, the DPA (art. 8.2) lets the Customer choose return or deletion.
Trains on customer data
No
Subprocessors disclosed
Yes
DPA available
Yes
GDPR contact

Hosting summary

Opper's platform is deployed in the AWS Stockholm region (eu-north-1), Sweden, and storage, traces and analytics stay in Sweden. Application data and files live in Amazon RDS, encrypted at rest with AWS KMS-managed keys; uploaded files go to a private S3 bucket where objects are segregated per organisation and encrypted with SSE-S3. Browser-to-platform and service-to-service traffic uses TLS across segregated VPCs and subnets. Backups run through AWS Backup and are encrypted, with daily snapshots kept five weeks and weekly snapshots fourteen months, restoration being reserved to Opper engineers. Internal access is authenticated through secured AWS accounts and SAML with Google, with 2FA or a passkey mandatory. The distinction that matters is between the platform and the inference route: inference itself can leave the EU depending on the route chosen, the EU routes cited being AWS Bedrock EU, Azure EU, GCP EU, Mistral and Berget AI. Sub-processors and their locations are published: AWS (Sweden), Auth0 (Germany, Ireland), Datadog (Germany), Google Workspace (Europe), Sentry (Germany) and Stripe (United States). Transfers outside the EEA are framed by standard contractual clauses, available on request.

Hosting countries
🇸🇪 Sweden
Hosting regions
EU
Watch-outs

Things to keep in mind

Risks and trade-offs to weigh before adopting Opper.

  • The total cost is driven by the model providers' rates, not by Opper: the 3% is only a platform fee, and one careless model choice can dwarf it
  • Moving to Control Plane raises the platform fee from 3% to 5.5%, and Observe runs a judge model that is billed in addition for every generation it scores
  • Full tracing stores prompts and responses; that is not the Gateway's default behaviour, so switching it on changes what leaves your control
  • Retention depends on the provider you route to, not on Opper alone, and must be checked route by route in the /models directory; Opper names one explicit exception on the provider side, DeepSeek's first-party API, which trains on the content sent to it
  • Choosing a non-EU route takes data out of the EU: residency follows the route, not the product, so a compliance commitment made in procurement can be undone by a one-line model change that nobody reviews
  • No independent security certification is published, and Stripe, listed as a sub-processor, is located in the United States
  • Read the contract against the pricing page rather than instead of it: the terms state that payments are non-cancellable and fees paid non-refundable, and they still name a Free Tier and a Utility Plan that are absent from the current commercial offer
Setup

Setup & Integrations

Technical difficulty

Low to moderate, but for developers only. The documentation's quickstart claims a working setup in under five minutes, and OpenAI SDK compatibility means an existing project needs only a new base URL and API key. The real prerequisites are modest: calling an API, handling an environment variable, and funding a credit balance. A coding agent can install the configuration for you (npx skills add opper-ai/opper-skills). There is no no-code interface, the product being aimed at developers. The control plane - routing, guardrails, budgets - is configured afterwards through rules, without redeployment.

Deployment

Web appAPI

Integrations

OpenAI Anthropic Google Gemini Mistral Meta DeepSeek XAI Alibaba AWS Bedrock Microsoft Azure Claude Code Codex Cursor Cline Aider Goose LibreChat OpenCode N8n LangChain Vercel AI SDK MCP Stripe Auth0 Datadog Sentry GitHub Discord
Company

Behind Opper

Company name
Opper Technology AB
Founded
29/08/2023
Country of origin
🇸🇪 Sweden
Headquarters
Västerängsvägen 51 A, 182 47 Enebyberg, Sweden
UBO
INFORMATION_NOT_FOUND
UBO country
INFORMATION_NOT_FOUND
Domain registrar country
🇺🇸 United States
Legal contact
Support contact

Fundraising

Pre-seed round of EUR 1.6 million announced on 18/04/2024, with Emblem, Luminar Ventures and business angels cited as investors - reported by the specialist press (swedishtechnews.com), not published by Opper
Second pre-seed round announced on 14/07/2025, with Luminar Ventures leading alongside Emblem, Greens Capital / Greens Ventures and business angels; the amount is disputed between sources, reported as EUR 2.5 million by EU-Startups, EUR 3 million by Vestbee and USD 3.0 million by TheSaaSNews, so it should be read as a range of roughly EUR 2.5-3 million rather than a settled figure; reported by the specialist press, not published by Opper itself

Social

Official links

Resources

All the official URLs gathered for verification and reference.

Compare

Alternatives

Tools that compete with or complement Opper.

O OpenRouterL LiteLLMP PortkeyV Vercel AI Gateway
FAQ

Frequently asked questions

Does Opper add a markup on tokens?
No. Model providers' rates are passed through at cost. What Opper charges is a platform fee taken when credits are purchased: 3% on Gateway, 5.5% on Control Plane, and custom terms on Enterprise.
Where is the data hosted?
The platform runs in the AWS Stockholm region (eu-north-1) in Sweden, and storage, traces and analytics stay there. Data remains in the EU unless you deliberately choose a route hosted outside it.
Does Opper train models on my data?
Opper states explicitly that it does not use customer data to train models. Provider-side policies are a separate matter and are published model by model in the /models directory and in the sub-processor list.
What does Opper store by default?
On Gateway, metadata only: model, tokens, latency, status and cost. Neither the prompt nor the response is retained. Storing prompts and responses requires the tracing available on Control Plane.
How do I migrate from OpenAI or Anthropic?
Through OpenAI SDK compatibility. Point your client at the base URL https://api.opper.ai/v3/compat, use your Opper API key, and pass the model as a provider/model string. The Anthropic, Gemini and Responses clients work against the same base URL.
Is there an API?
Yes. The API reference is published at docs.opper.ai/v3-api-reference and an OpenAPI specification is available, alongside Python and TypeScript SDKs, a CLI and MCP support.
Am I charged for failed requests?
No. Requests that fail, or that are routed to a fallback, are not billed. Only the call that succeeds is debited from the credit balance.
Is a data processing agreement available?
Yes. A DPA is published and dated 2026-03-14, together with a sub-processor list dated 2026-08-27 that names each provider, its location, its retention window and whether it trains on content.
Is there a minimum age?
Yes. The privacy policy (art. 11) states that the service is not intended for anyone under 16.
What deployment options exist?
Four: the shared European platform, a dedicated instance, deployment in your own cloud, or deployment in your own datacentre.
Conclusion

Should you pick Opper?

Opper makes a coherent proposition: one European endpoint for 700+ models, a single processor covering all of them, and a fee structure with no markup on tokens. For a team in the EU that has to say where inference runs, self-serve data residency from the first call - without an Enterprise contract - is the real differentiator, and the single-processor argument genuinely removes the DPA amendment each new provider would otherwise trigger. The published legal set is unusually complete for a company this young: terms of service, a dated DPA, a dated sub-processor list, and a security overview that states, provider by provider, retention windows and whether a provider trains on content.

The reservations are just as clear. No SOC 2 or ISO 27001 certification is published, so assurance rests on self-declaration. There is no about page and no team page: the site says nothing about founders, headcount or history. Whether a permanent free plan exists cannot be established, since the terms name a Free Tier that the pricing page does not show while the documentation asks for credits before the first call. And because Opper's charge is a percentage, there is no entry price to compare: the final cost is set by the model providers, and the 3% becomes 5.5% as soon as you want tracing, guardrails and budgets, with Observe billed on top.

Beyond the shared European platform, Opper offers a dedicated instance, your own cloud or your own datacentre - a broad ladder for a company incorporated on 29/08/2023. The fair summary is that Opper is a well-documented, EU-anchored gateway whose compliance story is strong on paper and unaudited by a third party; buyers who need a certificate rather than a policy should ask for one before committing.