Has this been fixed in later betas? I'm running beta three and even though I can now login to my Claude account and download the Claude agent, Xcode intelligence fails to generate any response and the error returned says I need to use ChatGPT/Codex. Here is the output. Your request couldn't be completed. Please authenticate with Codex by going to Settings → Intelligence → Codex → Account.

The accepted answer solved my issue (thanks again, Quinn). However, I would like to second the motion that those of us writing system extensions for iPadOS should NOT have to forego code coverage. Will there be an eventual fix for this?

Thank you Kevin and ssmith_c. It seems like there is a lot to unpack here. Let me see if I can short-circuit this discussion. Here is what I think is being suggested; Kevin, please correct me if I'm wrong or this will not work. Create an app owned by our organization that contains the shared DEXT. Publish this app in the AppStore. Sell the SDK with our XCFrameworks that talk to the DEXT. The customer needs to have the proper entitlements to talk to the DEXT and must also link in our SDK frameworks, which handle our IP and protocols for talking to the device. This means our SDK App is a simple DEXT delivery vehicle, which must be installed in order for any third-party application built with our SDK to have access to the DEXT. With this arrangement, the customer can build as many applications using our SDK as they like as long as only one of them is installed and operating at a time. Handling multiple apps installed and trying to communicate with our DEXT is out of scope for the time being.

@Kevin Elliott So, are you recommending that I build a debug and release version of the DEXT and make them an include dependency of the debug and release build process for the app that relies on the DEXT, sort of like I'm doing with the three key XCFrameworks that are also being included in the final application? If so, I would only rebuild and release the driver when it actually needed updating, yes? That would mean I could make the DEXT a part of the SDK workspace that builds the three XCFrameworks and have all of the build products dropped into a common build-output directory. Currently, I'm taking those frameworks and explicitly updating my application project file to include them, every time I have a new SDK release. If I understand you correctly, this might actually be the better approach for my case.

I've been told that the internal processing format of AVFoundation is always 32-bit float and that the buffers I'm handing to the write(from:) function must be in that format. Using a settings dictionary, Apple will automatically convert the 32-bit floats to the format in the settings dict. let settings: [String: Any] = [ AVFormatIDKey: kAudioFormatLinearPCM, AVSampleRateKey: sampleRate, AVNumberOfChannelsKey: channels, AVLinearPCMBitDepthKey: 16, AVLinearPCMIsFloatKey: false, AVLinearPCMIsBigEndianKey: false, AVLinearPCMIsNonInterleaved: false ] let audioFile = try AVAudioFile(forWriting: url, settings: settings) let format = audioFile.processingFormat I never pasted the code I use to create the audio file but the settings dictionary is supposed to guarantee the conversion I need so that 16-bit samples are indeed written to the disk. I've not tried this because I moved beyond this issue months ago by bypassing AVFoundation and writing the data directly to disk myself. The alternative seems to be to convert from the third-party format of Int16 interleaved buffers to Float32 non-interleaved buffers and then back again when writing to disk. If you need to guarantee a lossless solution, I don't think this is it, though Apple may disagree with me. If anyone wants to correct my thinking or perhaps contradict some bad information reflected in the post, please do.

I think I have a related problem in my code. Please see this link: https://developer.apple.com/forums/thread/822717 I'd like to understand better how your solution works and how you are using lockForConfiguration as a hardware readiness check.

I re-wrote the code after much experimentation, some time ago, and solved the issue. Please ignore.

I am seeing these errors as well. In between each session, I completely release and re-create the capture session to make sure I'm not holding onto any resources but it doesn't make a difference. What is the common wisdom on how to deal with this apparent OS issue? Is it advised to keep the original capture session around for the lifetime of the application? We currently do not do this because our camera is a USB device that may be swapped out for different models with different data formats, etc. In this case, the user has to stop and re-start our application in order to proceed, which is not ideal.

Quinn, I met you face-to-face back in 2005. Glad you are still around.

I've gotten OK, Bad-Gateway, Gateway Time-Out on four different systems. Even when OK is returned, it takes a while to get a response. This started almost exactly at 10am pacific time for me. Like many of you I used the security command line to check my certs, checked my firewall (PFSense), and even switch my build system and my target iPad to use my iPhone personal hot-spot, in order to bypass my infrastructure and still, no-joy. After reading your posts, I have become convinced that this is an Apple services issue, even though their status page says everything is working. More information. I have another project (Project-B) I'm working on, which is being developed on a different machine. Project-B installs on the same target iPad with no issues. I migrated the failing project (Project-A) over to this machine and added the missing certificates to the keychain so that the failing project would build. Even on this second system, I'm consistently getting the error for project-a but not for project-b. So, apparently this failure is specific to the type of certificate or provisioning for the target app, it seams. Project-A: Hardware app using DriverKit, custom USB hardware, microphone, camera, etc. Only viable on iPadOS with an M-Processor. Project-B: SwiftUI, SwiftData, document-based. Does use a FLIR camera but does not require DriverKit. Also uses microphone and camera. Are any of you using DriverKit or working with anything that corresponds to Project-A? Have you tried build a simple sample app that does nothing special; does it work?

The call to setVoiceProcessingEnabled(isEnabled: Bool) is synchronous, it triggers a configuration change in the engine, so you must ensure that you access formats from input and output node after changing the voice processing enabled state to ensure you’re configuring your graph with the correct formats. Additionally there is some sample code available here as part of the AVEchoTouch project - https://developer.apple.com/documentation/avfaudio/using-voice-processing?language=objc

From what I have observed, downsampling or upsampling is automatic, if you use a mixer. Specifically what I observed was that one end of the mixer is attached to the input node, which was running at 44.1KHz. The other end of the mixer was connected to an AVAudioPlayerNode running at 48KHz. I could hear the network transmitted audio coming out of the speaker. This confused me at first until I looked at the specific documentation for AVAudioMixer: "The mixer accepts input at any sample rate and efficiently combines sample rate conversions. It also accepts any channel count and correctly upmixes or downmixes to the output channel count."

@milesegan Were there any memory management issues switching from AVAudioPlayerNode to AVAudioSourceNode? I'm using the player node now and am having issues when the audio engine goes through a configuration change. When this happens I stop the engine, remove the player node(s), re-attach and re-connect the player nodes, and then restart the engine. I wrote this code before realizing source nodes were a thing. I'm hoping that using a source mode makes things simpler and require less dynamic coordination. My thinking is that I can have the requisite number of source nodes connected to a mixer and just leave that configuration around for the duration of my app. Then, when one of my two or three dedicated inputs comes online, I can feed buffers into the source nodes and not worry about adding and removing player nodes. From your experience, does this sound like it would work? Would you be willing to share some code showing how you configure your engine with the source node?

In my experience, things only consistently work when using Float32 non-interleaved samples. This seems to be the requirement for the audio engine input and output nodes as well asl playing back audio with the player node. I am also recording data to the disk in this format. Any time I tried to use Int16 interleaved data, the API results were negative. I had to perform my own conversions to and from these two formats because the third-party library I was using for remote-conference audio only accepted Int16 interleaved data in both directions.

When I had this problem, I seem to recall that I had to add the .allowBluetoothA2DP option to get my AirPods to work.