LLM-cpp: 26 Single-Header C++17 Libraries for Native LLM Integration

LLM-cpp introduces a comprehensive suite of 26 single-header C++17 libraries, meticulously designed for integrating large language models (LLMs) directly into native applications. The core philosophy behind LLM-cpp is simplicity and independence: each library is a self-contained `.hpp` file, allowing developers to drop in only what's needed without requiring Python, SDKs, or external package managers. This approach significantly streamlines the development process for C++ environments.

The libraries cover a broad spectrum of LLM-related functionalities, catering to various development needs. For instance, `llm-stream` facilitates streaming OpenAI and Anthropic responses via Server-Sent Events (SSE), while `llm-chat` manages multi-turn conversations. Developers building Retrieval-Augmented Generation (RAG) systems can leverage `llm-rag`, `llm-embed` for text embeddings, and `llm-rank` for passage reranking. The suite also includes critical tools for production environments, such as `llm-log` for structured JSONL logging, `llm-trace` for RAII span tracing, and `llm-cost` for token counting and cost estimation.

Further enhancing its utility, LLM-cpp offers libraries for advanced features like `llm-format` for JSON schema enforcement, `llm-guard` for offline PII detection and prompt injection scoring, and `llm-agent` for tool-calling agent loops. Many libraries boast zero dependencies, while others, particularly those interacting with external APIs, rely on `libcurl`. The project emphasizes natural composition, allowing developers to combine libraries like `llm-log`, `llm-retry`, and `llm-stream` to create robust, production-ready patterns. This native C++ solution promises to unlock new levels of performance and control for AI-powered applications, fostering innovation in areas where low-latency and minimal overhead are paramount.