
LlamaIndex is an open-source framework for developers building AI agents over their own data. It combines retrieval-augmented generation (RAG), tools, memory, and workflows for applications such as internal knowledge assistants.
A query engine can become one of an agent's tools, connecting document retrieval to actions. Developers can customize connectors, indexes, retrieval, and model integrations, while Workflows controls how steps interact.
Load documents, build an index, and expose it through a query engine. In Python, FunctionAgent uses tools such as ordinary functions or query engines; AgentWorkflow coordinates multiple agents. Model integrations include hosted providers and local options such as Ollama.
Workflows connects steps through events, with loops, branching, parallel paths, and human review. The framework can use LlamaParse for complex document processing, but that hosted service has separate pricing. Open-source access does not include unlimited hosted parsing or model usage.
Phil Mui of Salesforce praises retrieval customization and community support. Agent reliability still depends on the model; smaller local models can be less reliable. The maintainers now prioritize LlamaParse, a product direction to consider when planning framework adoption.
The separate hosted platform is now called LlamaParse, formerly LlamaCloud. Its US plans cover document services:
Choose the framework for control over your application; add hosted services when their document-processing capabilities fit your workload.
The framework is free, open source software you run yourself. Model APIs and hosting may cost extra; LlamaParse has separate plans.
Developers use it to build AI agents and RAG applications that retrieve context from their own documents, databases, and APIs.
LlamaIndex is the open source framework for RAG and agents. LlamaParse is a separate commercial platform for document parsing, extraction, and indexing.
+2 more
+2 more
+3 more