Klee

Klee is a native macOS AI chat app and agent for people who want an assistant to work with files on their Mac. It runs models locally through MLX, and downloaded models can chat offline without an account or API key.

• File and command tools: The assistant can read, create, overwrite and delete files, list folders and run shell commands.
• Model selection: Klee suggests models based on your Mac's memory and manages downloads, including resuming interrupted transfers.
• Image input: Supported Qwen 3.5 vision models accept images alongside text messages.
• Web access: The assistant can fetch webpage content; optional web search uses the external Jina AI service and requires internet access and a Jina API key.
• Response display: Answers appear as they're generated, with model reasoning in a collapsible card.

Supported models include Qwen 3.5, Qwen 3, Gemma 3 and DeepSeek R1, using 4-bit variants from mlx-community on Hugging Face. Inference and conversations stay on the Mac; web tools need network access.

Klee uses SwiftUI and doesn't require Docker or background services. It requires macOS Sequoia or later, an M1 or newer Apple Silicon chip and at least 16 GB of RAM. Larger models need more memory: Qwen 3.5 27B requires 32 GB, while Qwen 3.5 122B requires 96 GB.