Scaleway Generative APIs

Scaleway Generative APIs gives you access to popular AI models through simple, serverless endpoints. It's for developers who want to integrate text, code, vision, and audio models into their applications without deploying or managing any hardware. The service is a European alternative to US-based providers like Amazon Bedrock and Azure OpenAI.

A key part of the service is its focus on data privacy and sovereignty. All models are hosted in secure data centers in France. Scaleway states that it does not collect, read, or reuse your prompts or the model's outputs, ensuring your data remains private and compliant with European regulations.

The platform includes several features for building AI applications:

• OpenAI-compatible API: It's designed as a drop-in replacement for OpenAI APIs. This lets you use existing libraries and SDKs like LangChain with minimal changes.
• Pay-per-token billing: You pay for what you use, with pricing based on millions of input and output tokens. This model avoids fixed infrastructure costs. Batch APIs offer discounts for non-realtime tasks.
• Broad model access: Provides a selection of open-weight models for different tasks, including chat, code generation, image understanding, audio transcription, and creating embeddings.
• Structured data: A built-in JSON mode can force model outputs into a reliable, machine-readable format for easier integration with other systems.
• Function calling: Connect models to external tools and APIs, allowing them to perform actions and interact with other services to complete complex, multi-step tasks.

The serverless architecture means you can start testing models in minutes. For users in Europe, the service provides low latency with a fast time-to-first-token that's suitable for interactive chatbots and agent workflows.

This serverless offering is different from Scaleway's Managed Inference product. Generative APIs is best for getting started quickly with a pay-as-you-go model, while Managed Inference is for production use cases that require guaranteed throughput on dedicated, private infrastructure.