RecurseChat

Designed as a daily driver for Mac users who want the capabilities of AI assistants like ChatGPT without the privacy trade-offs of cloud processing, RecurseChat provides a native macOS experience built around local AI model execution.

The application's local AI model support includes Meta Llama 3.2, Llama 3.1, Gemma 2, Mistral, Llava (for vision tasks), and WizardCoder Python models that run directly on the Mac's neural processing hardware without internet connectivity.

For users who prefer cloud model quality for specific tasks, RecurseChat also connects to GPT-4, GPT-4o, Claude 3.5, and Claude 3 via API key providing the flexibility to choose local privacy or cloud capability depending on the sensitivity and complexity of each conversation.

Chat with PDF, Markdown, and text documents is one of RecurseChat's most practical features allowing users to upload their own documents and ask questions, request summaries, extract specific information, and engage in extended analysis conversations grounded in document content.

This local document chat capability is valuable for professionals working with confidential materials contracts, medical records, financial statements, proprietary research where document content must never be transmitted to cloud AI processing servers.

RecurseChat's interface is designed for the organization and management of extensive AI conversation histories supporting bookmarks, folder organization with nested folder structures, and advanced text search across all conversation content.

The ability to switch between multiple AI models within a single chat session gives users flexibility to leverage different model strengths at different points in a conversation using a fast local model for initial exploration and a more capable model for nuanced analysis within the same thread.