We've exposed AWS CloudFront headers in the request history tab. This gives you some additional context about where your app's requests are originating from.
AIProxySwift 0.135.0 is out, and it includes an overhaul of the Anthropic client. The rework makes it possible for us to expose much more Anthropic functionality.
There is a migration guide here to assist in updating your call sites: aiproxy.com/docs/swift-examp…
AIProxySwift is now swift 6.2 compatible! This is obviously a big change. If you upgrade to 0.130.0 and hit any weirdness, my DMs and the support chat in developer dashboard are open
AIProxySwift release 0.125.0 has GPT-5 params to control reasoning effort, text verbosity, and prompt templates
See readme sections:
"How to make a basic request using OpenAI's Responses API"
and
"How to use prompt templates with the OpenAI Responses API"
H/T to @Matt1Corey
You may get some deprecation notices the next time you update AIProxySwift. I've added a little blurb to the notice to guide you on what to add.
I'm making secondsToWait a required argument. It's up to the caller to provide a sensible timeout for the use-case / app / UX
I added a bunch of OpenAI Responses API examples to our README, including streaming vector store search and web search
These snippets are designed to be copy-pasteable for easy adoption in your app
The OpenAI Realtime SDK by @AIProxyTeam is now generally available.
It works equally well across Apple's platforms, with or without headphones.
The source for this demo is also available. This is the most functional realtime SDK on the market. It's MIT licensed. Have fun!
If you use our stableID feature, then updating to 0.105.0 will attempt to use StoreKit's appTransactionID as the identifier that we rate limit on. This is so handy.
Thinking of a future update where all requests require a verified appTransactionID. Will try with our apps first
We now have multi-speaker Gemini TTS support out of the box in AIProxySwift 0.103.0
Copy-pasteable snippets in the readme.
You can have a multi-speaker TTS app in approx 5 minutes
@AIProxyTeam now supports stableID. This ID will persist even if the user uninstalls the app, as long as they log in with the same iCloud account. It's useful for identifying users and can also be used as the RevenueCat User ID.
Editing images using gpt-image-1 is available in AIProxySwift 0.94.0.
The sample snippet below shows how to transfer the material from one image to another. It works really well.
I specified a `.low` quality in the snippet for speed of generation. Be sure to adjust for your use case
Full snippet: github.com/lzell/AIProxySwif…
Community contributions in latest release:
- Pass instructions to gpt-4o-mini-tts by Jordan Van Eetveldt
- Pass thinkingConfig to Gemini 2.5 flash thanks to @cristicrtu
- Support for @elevenlabs speech-to-text
I'll be running a workshop on the Responses API in Swift using @louzell_ 's AIProxySwift library if you'd like to join: tryswift.co/ai. But I agree that for most cases, the chat completions API is better for many reasons and should be kept around as the standard. However, if you're using an OpenAI model specifically and have an agentic system and one of your agents needs to use a specific OpenAI model (like o1), it's fine to use.
If you are feeling inspired to build style transformation apps, search for "How to edit an image with Gemini" in our readme. You can copy and paste the snippet straight into your Xcode project. Ship fast.
Have fun!