OpenAI-Compatible API on EU-Sovereign Models: Usage-Based Pricing Explained

In one sentence: Linkworld's Platform API is an OpenAI-compatible endpoint that routes calls to a menu of EU-sovereign open-weight models plus frontier models through the same router, meters usage per tenant instead of charging by seat, and adds Managed RAG (embeddings, storage, and queries) for teams that don't want to run their own retrieval pipeline — currently in Early Access.

If you're a SaaS founder or CTO evaluating where to run your product's LLM calls, you've likely run into the same fork every EU-based team hits: stay on a US-hosted API for convenience, or move to EU-sovereign infrastructure and rewrite your integration around a different client library, different auth model, and a pricing structure you have to reverse-engineer from a sales call. Linkworld's Platform API is built to remove that trade-off: the interface is OpenAI-compatible, so an integration built against it doesn't need a separate SDK or a different request shape — only the base URL and the model name change — while the models it routes to run on EU-sovereign infrastructure.

What "OpenAI-Compatible" Actually Buys You

For a SaaS team, "OpenAI-compatible" is a migration-cost claim, not a marketing label. If your product already calls a chat-completions-style endpoint, pointing it at Linkworld's Platform API is a base-URL and model-name change, not a rewrite of your request handling, streaming logic, or error handling. That matters most when you're testing whether an EU-sovereign model can replace or sit alongside a US-hosted one in production: you can swap the target and compare quality, latency, and cost without re-architecting the calling code around a new API shape first.

The EU-Sovereign Model Menu, Plus Frontier Through the Same Router

The Platform API doesn't lock you into a single model. It fronts a menu of EU-sovereign, open-weight models — including gpt-oss, Qwen 3.5 and 3.6, GLM 5.2, Llama, and Mistral — so you can pick the model that fits a given task's cost and quality bar, and switch between them without touching your integration code. For work that needs a frontier-class model rather than an open-weight one, frontier models are reachable through that same router, so you're not standing up a second, separately-integrated API just to reach for a stronger model occasionally.

Usage-Based Pricing, Metered Per Tenant

The Platform API is priced on usage, not seats: you pay for what you actually consume per tenant, not a flat per-user license regardless of how much your product calls the API. That's the pricing shape most SaaS teams want from infrastructure they're embedding into their own product — cost that scales with your customers' actual usage, not a fixed headcount-based bill that's disconnected from it.

Managed RAG Without Building Your Own Pipeline

Retrieval-augmented generation is usually the part of an LLM integration that takes the longest to get right: embedding generation, vector storage, and query-time retrieval each need their own infrastructure decisions. Linkworld's Managed RAG handles embeddings, storage, and queries as part of the same platform, billed by usage rather than requiring you to provision and operate a separate vector database and embedding pipeline yourself.

What Early Access Means Today

The Platform API is in Early Access. Self-service API-key issuance isn't productized yet, so if you're evaluating it for a production integration, expect an onboarding conversation before you get credentials rather than a self-serve signup flow you complete in a browser tab. That's a deliberate, current-state limitation, not a hidden catch — it's worth knowing before you plan an integration timeline around it.

Why This Sits Under the Same Governed Loop

The Platform API isn't a separate product bolted onto Linkworld — it's the same multi-LLM infrastructure that runs Linkworld's own governed automation engine, the Plan → Debate → Execute → Review → Assess loop that plans and verifies work against actual system state rather than an agent's own report. If you're building your own agentic features on top of the Platform API, you're building on the same routing and model infrastructure that underlies a production governed-AI system, not a lighter-weight side project.

Frequently Asked Questions

Is the Platform API a drop-in replacement for the OpenAI API?

The interface is OpenAI-compatible, so an existing integration typically needs only a base-URL and model-name change to point at it — not a rewrite of your request or response handling.

Which models can I call through the Platform API?

A menu of EU-sovereign, open-weight models — including gpt-oss, Qwen 3.5 and 3.6, GLM 5.2, Llama, and Mistral — plus frontier models reachable through the same router.

How is the Platform API priced?

Usage-based, metered per tenant — you pay for what you actually use rather than a flat per-seat license.

Can I get an API key today?

The Platform API is in Early Access. Self-service key issuance isn't productized yet, so access currently starts with an onboarding conversation rather than an instant signup.

Does the Platform API include retrieval (RAG)?

Yes — Managed RAG covers embeddings, storage, and queries as part of the same platform, billed by usage, so you don't need to run your own vector database and embedding pipeline.