github.devotd.
ReleaseModel Context Protocol

r44vc0rp/isocast v0.0.7 — Editing project export and GPU sharing

isocast v0.0.7, the OBS alternative for Apple silicon Macs, adds project export for editing programs and shared graphics device use.

Repositoryr44vc0rp/isocast ↗

Free, open-source OBS alternative for Mac: scenes and sources, every source recorded as its own file (ISO), about half OBS's CPU, CLI and MCP control. Rust, Apple silicon.

stars
★ 93
Forks
5
Language
Rust
License
Apache-2.0
Latest release
v0.0.7
Read original source ↗View on GitHub ↗

isocast is a free, open-source OBS alternative built exclusively for Apple silicon Macs. v0.0.7, released on October 18, 2026, puts editing project export front and center, handing off to Final Cut Pro, Premiere Pro, and DaVinci Resolve. In the exported project each angle lands on its own track and points at the original file, so no re-encoding or copying is needed. Trim ranges, markers, and the original audio come along too. During playback at 2× and 4×, pitch-corrected audio plays in sync with the picture.

This export keeps you working on the recorded material inside an editing program instead of rebuilding it as separate files. isocast divides the composition into scenes and sources and records every source as its own ISO file. The repository description says CPU usage is around half that of OBS and that it can be driven from the CLI and via MCP. In this version the whole app now shares a single graphics device. The release notes report that GPU time and memory roughly halved when several app instances were running. It also supports a playback buffer and virtual cameras.

That said, the release notes given here do not state the measurement conditions or comparison method behind the performance numbers. The CPU usage figure is likewise a claim from the repository description, with no separate measurement backing it. Support is also limited to Apple silicon Macs. So what can be confirmed is recording on this machine and the editing hand-off features; there is no basis for generalizing the performance to other environments or workloads.