AIProxySwift now has broad support for OpenAI's Responses and Conversations APIs. You can build an assistant that remembers prior turns without repeatedly sending its entire history from the app, while keeping the same protected-key setup you use for other OpenAI requests.

Two ways to carry context forward

For a lightweight conversation, pass the ID of one response into the next request with previousResponseId. For longer-lived state, use the Conversations API to create a conversation and manage its items independently of any single response.

AIProxySwift includes typed methods to create, retrieve, update, and delete conversations; add, list, retrieve, and delete conversation items; and paginate through a conversation's history.

More of the Responses API, represented in Swift

Responses can be returned all at once or consumed as an async stream. The library also represents the richer items the API can produce, including reasoning, citations, web and file search, image generation, code interpreter output, computer actions, shell tools, custom tools, MCP calls, and function calls.

Recent additions also cover response instructions, storage controls, reasoning effort through xhigh, maximum output tokens, usage details, and context compaction. The goal is to let your app use OpenAI's provider-specific capabilities without dropping down to handwritten request and response types.

Start building

The OpenAI section of the AIProxySwift README contains complete examples for basic and streaming Responses requests, tool calls, and conversation management. AIProxySwift is MIT licensed and supports both direct API keys during development and protected AIProxy services for production.