Arbiter – Local AI Chat

About this app

Arbiter is a privacy-first offline AI assistant that runs on your device. Choose from lightweight open-source models like Gemma, Llama, DeepSeek, Qwen, and Mistral, or connect to larger models running on your PC or Mac over your local network. You can also connect to Arbiter for Mac when it is serving a local model on your network. There are no accounts, servers, or data collection. All your conversations are stored securely on your device, under your control.

Key Features:

– On-Device AI: Download and run models locally on your iPhone
– Apple Intelligence Support: Use Apple’s on-device Foundation Model on supported devices without downloading another model
– Local Image Generation: Create images directly on your device with Stable Diffusion
– Local Network Models: Connect to more powerful models running on your PC or Mac through tools like LM Studio, Ollama, or Arbiter for Mac
– Visual AI Support: Upload images or take photos in-app to use with MLX-powered visual models
– Document & File Uploads: Summarize files and extract key insights
– Expanded File Support: Work with DOCX, RTF, HTML, Markdown, CSV, JSON, code, PDF, and text files
– Bring Your Own Model: Search Hugging Face, download compatible models, or import supported GGUF files and MLX folders already on your device
– MCP Tools: Connect to compatible MCP servers, review tool arguments, and run tools when needed
– Optional Web Search: Access up-to-date information when needed, while keeping your core experience private and local
– Smart Chat Suggestions: Context-aware suggestions tailored to your selected model and role
– Personalization: Customize system prompts and model settings for a more tailored experience
– Advanced Model Controls: Adjust context size, maximum response length, temperature, Top-P, and Top-K
– Local Model Benchmarking: Measure model quality, speed, response time, and memory use directly on your device
– Privacy by Design: Your conversations and data never leave your device unless you choose to enable web search or connect to a local network model
– Model Flexibility: Choose from multiple open-source AI models to balance speed, accuracy, and memory use
– Improved Reliability: Smarter context handling and enhanced error management for a more stable experience
– No Hidden Costs: Avoid expensive cloud fees with local and local-network processing

Whether you want to brainstorm, code, proofread text, analyze images, summarize files, search the web, or experiment with the latest open-source AI, Arbiter makes advanced AI models accessible while keeping your data in your hands.

Device Requirements: For smooth on-device performance, Arbiter is recommended for iPhone 13 Pro or later, or devices with at least 6GB RAM. Older devices may experience memory issues or crashes.