LangChain Swift is a pure client-side Swift library for building applications powered by large language models (LLMs) and AI. It provides a comprehensive toolkit for developers targeting iOS, macOS, and visionOS, enabling seamless integration of LLM capabilities directly within native applications without requiring a backend server.

How It Works

The library implements the core LangChain concepts, including LLMs, prompt templates, memory management, document loaders, text splitters, vector stores, retrievers, agents, and output parsers. It offers a modular architecture allowing developers to chain components together to create complex AI workflows. The library supports various LLM providers and local model inference, with a focus on efficient execution on Apple platforms, including Metal acceleration for local models.

Quick Start & Requirements

• Install: Add https://github.com/buhe/langchain-swift to your Swift Package Manager dependencies. For local models, use .branch("local").
• Prerequisites: API keys for various services (OpenAI, Supabase, Serper, etc.) are required. For local models, specific model files are needed. Metal support is available for GPU acceleration.
• Setup: Configuration involves setting environment variables or using LC.initSet() with API keys and other necessary parameters.
• Docs: bilibili-api

Highlighted Details

• Supports a wide range of LLM providers including OpenAI, Hugging Face, Llama 2, Gemini, and local models via Ollama and LMStudio APIs.
• Integrates with Supabase for vector storage and retrieval.
• Includes tools for web scraping (via Serper), location, weather, and text-to-speech.
• Features agent and chain capabilities for complex task execution and conversational AI.
• Offers various output parsers for structured data extraction from LLM responses.

Maintenance & Community

• Active development with a clear roadmap covering LLMs, vector stores, chains, agents, and more.
• Community support available via Slack: https://join.slack.com/t/langchain-mobile/shared_invite/zt-26tzdzb2u-8RnP7hDQz~MWMg8EeIu0lQ
• Open to contributions for new features, infrastructure improvements, and documentation.

Licensing & Compatibility

• The repository does not explicitly state a license in the README.

Limitations & Caveats

• The library is currently in beta.
• Some features, like Distilbert embeddings, are marked as not yet implemented on the roadmap.
• The README mentions watchOS (part), indicating potential limitations for that platform.