Circleback.ai Rebuilds Recording Engine in Swift to Bypass Electron Bottlenecks
The company offloaded performance-critical capture tasks to a native layer to eliminate garbage collection pauses and throttling.
Circleback.ai has migrated its meeting-recording engine from the Electron render process to a native Swift implementation. The move aims to eliminate reliability issues inherent in browser-based runtimes when handling high-performance, real-time data.
To achieve this, the company rebuilt the engine to operate outside the Electron render process, which Circleback.ai Engineering noted is the "wrong place to do realtime audio and video capture." According to the company, capture engines cannot tolerate the garbage collection (GC) pauses or thread throttling that browser runtimes use to maintain responsiveness. The entire rewrite was completed and shipped within two months.
A Native Approach to Capture
The new architecture utilizes platform-specific APIs to ensure stability. On macOS, the engine employs ScreenCaptureKit and manages clock synchronization across three independent capture sources. For Windows, Circleback.ai implemented a custom wrapper called OBSKit to utilize libobs and Windows Graphics Capture (WGC), maintaining a fallback to BitBlt for frame delivery.
To protect against data loss during potential system crashes, the company shifted from standard MP4 files to fragmented MP4 (fMP4). This format ensures that recorded data is saved in smaller, independent segments rather than relying on a single finalization step at the end of the recording.
Bridging Swift and React
While the heavy lifting now occurs in a native layer, Circleback.ai maintained its React-based user interface. To connect the two, the team developed a custom bridge called "Atomic." This tool leverages the @NodeExport macro to automatically map Swift @Published properties to Jotai atoms within React, allowing the UI to react to native state changes in real time.
Industry Implications
This migration highlights a growing trend among Electron developers to offload performance-critical systems to native code. By moving audio and video capture to Swift and C-based libraries, Circleback.ai bypassed the jitter and crashes often associated with the JavaScript event loop. The implementation of the Atomic bridge is particularly significant, as it reduces the amount of manual "glue code" and C++ bindings typically required to synchronize a native runtime with a web framework.
Looking Ahead
The success of this native migration provides a blueprint for other cross-platform applications struggling with real-time media processing. As Electron apps continue to handle more complex system-level tasks, the use of reactive bridges to connect native performance with web-based flexibility is likely to become a standard architectural pattern for high-reliability software.