A local-first multi-agent desktop application: hand your goal to the "Commander," which either completes it itself or delegates it to 16 built-in specialists (deep research, writing, PPT, images, video, voiceover, UI, programming, Office, e-commerce, etc.), and can also orchestrate locally installed agents like Claude Code, Codex, and OpenCode as part of the team.
Star count 4908 (+61 / 24h)
Reasons
- The project saves goals, rules, memories, and files, allowing new conversations to continue where the previous one left off; to-do items can be assigned to humans or agents, and when auto-advance is enabled, the team automatically picks up the next task; supports scheduled tasks ranging from hourly to monthly
- VideoStudio can reference the structure, style, and pacing of a sample video to recreate a promotional video using your own product assets; connectors include Gmail, Google Calendar, Notion, Slack, Shopify, Stripe, etc., with confirmation required by default before sending, publishing, or deleting
- Models can be chosen from built-in Orkas models, your own API keys, or local endpoints; when using your own keys, model calls are sent directly from your local machine to the service provider
Usage
- The packaged version supports macOS 12+ and Windows 10+, with 500 credits granted upon login; for the source code version: run ./run.sh after git clone (use run.cmd on Windows), requires Node 22.12+, and the first launch will download Python, FFmpeg, whisper.cpp, and an embedding model of approximately 95 MB
- Linux only supports the source code version and requires glibc 2.34+
Risks
- Shared projects, cloud sync, and app publishing are only available in the packaged version and are not supported in the open-source version; the packaged version requires login, built-in models are billed by credits, and voice input uses Orkas's recognition service
- The README focuses primarily on product introduction, with no independent evaluation of each specialist's performance; the repository was created in 2026 4, licensed under MIT