Instant mode audio not captured when recording multiple videos

[smcjustin]

Description

Some videos recorded in instant mode were missing audio after upload. I double-checked system permissions and recording settings but could not figure out the exact cause. Eventually I realized it was only successfully recording audio on the first video after opening the app. All the videos after that were missing audio. Quitting and reopening Cap would allow one successful upload and then go back to videos with missing audio.

I believe in all cases the .cap recording files are missing the /audio folder. I am not sure if the same issue affects studio mode.

Reproduction

I'm guessing this bug may not be universal and could be platform-specific otherwise I would expect it to have already been reported. But here are the steps to reproduce on my machine:

1. Quit Cap if already open and then Open Cap
2. Record a video in instant mode and upload it
3. Confirm video uploaded successfully with audio
4. Return to the Cap window and record a new video without changing any settings
5. Video uploads without sound

Additional Context

• Cap version: 0.4.84
• Operating system, version: MacOS 26.4.1
• Device (optional): MacBook Air 15" M4 2025

cap-desktop.log.2026-05-14.txt

On a personal note, I have to re-record hours of training videos because of this bug. I am paying for the "pro" version of Cap and to be honest it's pretty ballsy to be charging for a pro version when issues like this make it to production.

[linear]

CAP-709

[tembo]

Some context from the codebase that might help investigate this:

The instant mode audio pipeline uses a MicrophoneFeed actor that broadcasts samples to multiple MicrophoneFeedSender instances (see crates/recording/src/feeds/microphone.rs). Each recording session adds a new sender via AddRecordingSender, and disconnected senders are removed when samples are processed (lines 904-916).

The mic feed itself is reused across recordings - it's locked via lock_selected_microphone() in apps/desktop/src-tauri/src/recording.rs. This reuse pattern means there could be state leakage if the previous recording's sender cleanup doesn't complete before the new recording starts.

Worth noting that #1740 reports a potentially related symptom on the same version (v0.4.84, macOS) - audio waveform appears during recording but is missing afterward. The common thread might be the audio folder not being created properly or the sender not being wired up correctly on subsequent recordings.

The attached log file should be helpful for tracing exactly where the audio pipeline diverges between the first (working) and second (failing) recordings.