API
Application Programming Interface. A structured way for one piece of software to talk to another. In DeAI, APIs let applications request inference from a model without running the model themselves.
Also known as: application programming interface, REST API
Every modern AI product exposes an API. When you use ChatGPT on the web, the browser sends your message to OpenAI’s servers via an API, they run inference on their model, and they send the response back. When a developer builds an app that summarises documents or answers questions, they’re making API calls to whichever model provider they picked. The API is the contract: you send a structured request, you get a structured response, and everything in between is the provider’s problem.
APIs are what make DeAI practical. A hosted API is the simplest way to use a model without owning the hardware. You pay per request, per token, or via a subscription; the provider runs the GPUs and maintains the software. Almost every DeAI inference project offers an OpenAI-compatible API (Venice, Morpheus, io.net, Bittensor subnets) which means existing software written for OpenAI can switch providers by changing one URL and one API key. That compatibility is a deliberate strategic choice: it turns switching from a rewrite job into a config change, which is exactly what makes the inference market competitive.
The economic structure of an API is usually pay-per-use. You pay for input tokens (what you sent), output tokens (what the model generated), and sometimes a base rate per request. Token prices vary by model. Claude Opus, GPT-5, and similar frontier models cost 10-100x more per token than smaller open-weight models running on commodity hardware. This is the primary reason DeAI inference exists as a market: there’s real demand for lower-priced alternatives to frontier APIs, even at slightly lower quality.
The sovereignty angle on APIs is about trust and privacy. When you use an API, the provider sees your prompt. They can log it, use it for training, hand it to governments, or get breached and leak it. DeAI inference projects address this with various combinations of TEE-based privacy (Venice, Phala, NEAR AI), confidential compute (SGX, TDX enclaves), and encrypted routing. None of them are equivalent to running the model yourself, but they’re all materially better than a standard OpenAI API call from a privacy standpoint. The OYM Data Sovereignty dimension scores each project on how much the API actually hides from the operator.