Local Inference
Version 1.1 (58) · October 19, 2026
- macOS 14+. Apple silicon and Intel. Signed and notarized.
- Chat, image generation, video, speech, and branching Stories.
- Local models or supported providers with your own keys. Keys stay in Keychain; cloud providers receive submitted content and may charge.
- Model weights download separately. Local images and Kokoro speech require Apple silicon.
- Optional FLUX 3 Image Klein Q4 uses a separate runtime and 5.3 GB model download on Apple silicon with macOS 26+. Review the community conversion and original model license before use.
- Image and video galleries with resizing, selection, and deletion; images also support zoom and navigation.
- Half-private keeps saved work and blurs images. Private mode keeps new work transient; saved prompts remain available to reuse.
- Reveal All temporarily shows blurred images. Hide Images or backgrounding restores blurring.
- Image queues pause across app restarts. Local text and gallery upscaling have explicit setup.
- Local image generation uses calibrated memory estimates and retains work until stopping is confirmed.
- Optional rewriting after an image prompt refusal is off by default and keeps the selected image model.
- Stories preserves drafts and ignores actions from a previously selected story.