Overview

Dedicated APIs

• Routing looks at a prompt or an agent task and tells you how hard it is, so you know which model fits it.
• Inference takes a request in the same format as the Anthropic and OpenAI SDKs, picks a model for it, applies optimizations and returns the response. You could use Anthropic/OpenAI models, or open-source ones at better efficiency.
• Beacon looks at an agent's run and tells you if it's stuck or needs an intervention.

Keys and endpoints are on the API dashboard.

Existing tools

If you already use a CLI, editor, or agent framework, it can send its requests through Downsizing instead of straight to the provider. The tool works the same as before. See Integrations for setup per tool.

Provider subscriptions

If you already have an Anthropic or OpenAI subscription, you can connect it and use it through the Inference API. We usually see a 2-3x extension of your limits. Connect Anthropic · Connect OpenAI