LlamaIndex Newsletter 2024-02-20: Introducing LlamaCloud

LlamaCloud has been unveiled, presenting new production-grade context augmentation infrastructure for LLM and RAG applications.

The core components consist of two parts.

• LlamaParse: A specialized parsing service for handling complex documents containing tables and figures, strengthening support for semi-structured documents.
• Managed Ingestion and Retrieval API: Simplifies data loading, processing, and storage, reducing the operational burden of RAG pipelines.

The scope of integration is also broad. Through LlamaHub, it supports 150+ data sources and 40+ data stores, emphasizing scalability tailored to enterprise workflows.

The rollout stage differs by service. LlamaParse is currently in public preview, focusing first on PDFs, with usage limits for public users. Commercial terms require separate inquiry, and the Managed API is in private preview, available only to select enterprise partners.

Highlights from this newsletter were also compiled.

• Corrective RAG LlamaPack: Reflects ideas from the CRAG paper to improve the accuracy and relevance of retrieval results.
• SELF-DISCOVER LlamaPack: Implements a meta-reasoning-based problem-solving approach as a LlamaPack.
• RAG Production Guide: An operations-focused RAG guide introduced by Sisil Mehta of JasperAI has been released.

Numerous demos, tutorials, and guides were also introduced. These broadly cover DanswerAI's enterprise knowledge integration case, a multimodal app for ADU Planning, vector embedding setup, a research RAG agent, an introduction to LlamaIndex v0.10, building agents with Query Pipeline, building a multimodal app with Ollama, RAG evaluation, and a SageMaker deployment case.