Lemonade exits macOS beta with full omni-modal local AI stack
The open-source local AI runtime now ships image generation, speech synthesis, transcription, and OmniRouter on macOS alongside Linux and Windows.
Lemonade, an open-source local AI runtime, graduated macOS support out of beta this week. The platform now runs image generation, speech synthesis, transcription, and its OmniRouter orchestration layer natively on Apple hardware. Developer GeramyL led the port, bringing feature parity with the existing Linux and Windows builds.
Lemonade positions itself as a zero-telemetry alternative to LM Studio and Ollama. The project ships as a 3 MB portable binary with no cloud dependencies or upsell prompts. Users can input and output images and speech alongside text — a multimodal workflow the team calls "omni-modal."
What stands out
- 01Zero telemetry, zero cloud. The project commits to local-only operation with no analytics, no phone-home, and no premium tier nudges. The codebase is fully open on GitHub.
- 023 MB binary. The entire runtime compiles to a single portable executable under 3 megabytes. Developers can bundle it into other projects without dependency sprawl.
- 03OmniRouter. The platform includes a routing layer that can coordinate multiple local models — text, image, speech — in a single workflow. Prompts can trigger image generation or TTS without switching tools.
- 04Cross-platform from one codebase. The same source tree now builds for Linux, Windows, and macOS. An iPhone app is in development to access the same local stack over the network.
- 05Community-driven roadmap. The Discord server and GitHub issues drive feature priorities. No venture funding or commercial entity steers development.