Ka1zen icon
Latest release

Local AI for
Apple Silicon

Chat with open LLMs entirely offline on your Mac. Private by design, powered by Apple MLX. No account, no cloud, no telemetry.

macOS 15+ Apple Silicon M1+ 100% Offline ✨ Free forever
⬇ Download Ka1zen GitHub 📖 User Guide
0 €
No subscription
0 kb
Data sent to cloud
200+
Models supported
Everything in one app
Ka1zen chat interface
Onboarding prerequisites Model Manager General settings Model settings
Nothing leaves your Mac
💬

Local Chat

Qwen, Gemma, DeepSeek, Mistral, Llama and anything on mlx-community. Real-time streaming.

🎨

Image Generation

FLUX.2-klein-9B via mflux, locally. Say "draw a…" and Ka1zen generates. No API key needed.

🔍

Grounded Images

Ask Ka1zen to search the web AND generate an image. It fetches real facts, then draws from them.

🧠

Thinking Mode

Watch the model reason step by step. Works with Qwen3, DeepSeek-R1, Gemma 4, Nemotron.

🌐

Web Search

DuckDuckGo + direct page fetch. URL auto-detection. Clickable [1]-style source citations.

🖼

Vision

Drop images into the chat. Supports Gemma 4, Pixtral, Mistral Small VLM, Qwen2-VL, LLaVA.

📚

RAG

Index PDFs, text, and code locally with Apple NLEmbedding. Fully local SQLite storage.

🔌

API Relay

OpenAI-compatible proxy for Continue, Cursor, Zed, LM Studio — use local models anywhere.

🔎

Token Inspector

Per-token logprob visualisation — spot hallucinations with green / yellow / red colouring.

Search the web. Generate an image. One prompt.

Ka1zen fetches real facts first, then turns them into a FLUX-optimized prompt and generates locally.

1

Web Search

Ka1zen queries DuckDuckGo to gather real, up-to-date facts — visible as a live Agent Step with source citations.

2

Prompt Synthesis

The active LLM converts search snippets into a detailed FLUX-optimized English prompt, incorporating colors, shapes, era, and context.

3

Local FLUX Generation

The image is produced on-device with FLUX.2-klein-9B. No image ever leaves your Mac.

Example prompts

Search for a 1967 Ford Mustang and create a photorealistic image of this car parked in front of an American diner at sunset.

Look up the Renaissance fresco technique and make a detailed illustration of an artist at work on a church ceiling.

Find info about Japanese ramen shops in Tokyo and generate a cinematic photo of a steaming bowl of tonkotsu ramen.

Search the web for James Webb Space Telescope imagery and draw a stylized poster celebrating its deep-field observations.

What you need
macOS 15 Sequoiaor later
Apple SiliconM1, M2, M3, M4 or newer
16 GB RAM32 GB recommended
Python 3.14from python.org only
mlx-lmvia included install.sh
mfluxoptional — image generation
🔒

100% private. No account required. No analytics, no crash reports, no "anonymous usage statistics". Your conversations, documents and embeddings never leave your Mac.