LlamaIndex Newsletter 2024-02-13

LlamaIndex v0.10 has been released, restructuring the framework with a focus on production readiness rather than just productivity. The core package has been reorganized, and hundreds of integrations and LlamaPacks have been split out into separate PyPI packages. The existing LlamaHub has been reorganized as a central hub gathering all integrations, and ServiceContext has been deprecated for a better development experience.

At the same time, a short course on Advanced RAG Development was also introduced. It covers everything from unstructured data processing to agent integration, and is structured as hands-on practice using LlamaIndex Query Pipelines to learn from the basics of text-to-SQL to advanced query techniques.

This week's highlights were Self-RAG, the FlowiseAI integration, and MistralAI's RAG guide. Self-RAG is provided as a LlamaPack, making it easy to attach dynamic retrieval and iterative RAG workflows, while the FlowiseAI integration helps with drag-and-drop, no-code RAG app development. A LlamaIndex-based RAG guide using Mistral-medium and Mistral embedding models was added to the Mistral documentation.

Features and tools have also expanded. The RAG CLI lets you search files on the local file system using on-device embeddings, and create-llama spins up a full-stack agent server equipped with 50+ tools with a single CLI command. In addition, agent functionality was added to LlamaIndex.TS, and integration with DeepEval strengthened RAG evaluation and unit testing for CI/CD.

A variety of guides and tutorials are also available. These include a step-wise query guide for agentic RAG, a guide for integrating with Hugging Face's Messages API, building Query Pipelines, building RAG apps with Llama2 and LlamaIndex, knowledge memorization based on Mistral-7B fine-tuning, and an introduction to NeMo Guardrails. Finally, for enterprises, it asks for interest in LlamaIndex's enterprise-ready products and partner program.