A local assistant becomes more useful when it can do more with your ideas. In the 0.6.x releases for macOS and iOS, LocalEngine adds new ways to create images, classify information, choose models, and start a useful conversation. Here is what you can try, and where.
Turn a prompt into an image on your Mac
Image generation joins chat in LocalEngine for Mac. Download the free Qwen-Image model from Settings → Models → Image Generation, then open Engine → Image Generation or press ⌥⌘G. Describe a scene, a poster, or an illustration; Qwen-Image also supports rendering English and Chinese text within images.
Choose the image size, steps, guidance, and seed, or add a negative prompt. Follow the render progress, stop when needed, and use the gallery to copy, save, share, or reuse an image’s settings. Apps and scripts can also generate images through the local /v1/images/generations API. The model package is about 18 GB; allow enough storage and unified memory, especially on smaller Macs.
Get a decision your workflow can use
Some tasks need a clear choice rather than a conversation. Clef decision models answer yes/no, multiple-choice, and rating questions about text, JSON, or images, returning a probability for every option. Use them to route a request, triage incoming information, or classify content through the local /v1/systemone API, without parsing a prose answer.
Choose Clef Flash (free, about a 7 GB download, for Macs with 16 GB of memory) or Clef 27B (Pro, about 20 GB, requiring 32 GB or more). Select “Use for Decisions” to set the model; it loads when a request needs it. These models work through the API and stay out of the chat model menu.
Choose MLX models on iPhone and iPad
Qwen3.5, Gemma 4, MiniCPM5, and LFM2.5 gain MLX versions on iOS, alongside their GGUF versions. Download them from the Models tab and select one as usual: LocalEngine automatically switches to the matching engine, with no extra settings. Each MLX model keeps the same Free or Pro status as its GGUF counterpart.
Make room for Bonsai 2
Bonsai 2 brings ternary models with image understanding and tool calling. On Mac, Bonsai 2 27B (Pro) adds reasoning in a 7.6 GB model download, plus a 0.6 GB vision projector. It needs at least 16 GB of memory and may spend time thinking before it answers.
On iPhone and iPad, Bonsai 2 Small comes in 2B (free), 4B, and 9B (Pro) sizes. Downloads including the vision projector are about 0.9 GB, 1.6 GB, and 3.2 GB respectively; the 9B model is intended for devices with at least 8 GB of memory. These releases use GGUF models. The 27B model is Mac-only, while the Small collection is on iOS.
Start with a task, not an empty chat
New chats now offer example prompts for summaries, translation, rewriting, code explanations, email drafts, and meeting minutes. Open “All prompts” to explore 20 templates across writing, language, data, coding, and work. Fill in a short form and start the conversation. Templates are bundled offline and follow the app language in all 10 supported languages.
Long fields can be filled from text, Markdown, or PDFs with a text layer; Mac also supports Word and OpenDocument files. Scanned PDFs are not supported. For GGUF models, a warning before sending helps you spot input that exceeds the current model’s context window. On iOS, install a chat model first to enable the example prompts.
Try what’s new
To get started, update LocalEngine, open Models to explore the new collections, or try an example prompt in a new chat. Once the models are downloaded, these features run on your device. Your prompts, documents, and generated images stay local, with no account, no cloud processing, and no telemetry.