Llama Cloud Services

Llama Cloud Services is the Python client SDK for LlamaCloud, LlamaIndex’s hosted platform for document AI. It bundles three services behind a single package: LlamaParse, a GenAI-native document parser for complex PDFs and office documents; LlamaExtract, an agentic structured-data extractor that turns unstructured documents into typed JSON; and LlamaCloud Index, a managed, automated ingestion and retrieval pipeline for RAG applications.

The package is a thin, typed wrapper around LlamaCloud’s hosted API — installation and setup is a single pip install plus an API key, with EU-region endpoints available for data residency. It is widely used as the parsing/extraction layer in RAG and agentic document-processing pipelines built on top of LlamaIndex or standalone. Note: as of this writing the maintainers have marked this repository as deprecated in favor of a new llama-cloud package (run-llama/llama-cloud-py), though llama-cloud-services remains published and in active use.

What You Get

• LlamaParse client for GenAI-native parsing of complex PDFs, Word docs, PowerPoint, and other formats into markdown or structured output
• LlamaExtract client for turning unstructured documents into typed, schema-defined JSON via an agentic extraction pipeline
• LlamaCloudIndex client for a fully managed, automated document ingestion and retrieval pipeline for RAG
• Built-in EU region support (EU_BASE_URL) for teams with data-residency requirements
• A CLI (llama_cloud_services/parse/cli) for parsing documents from the command line without writing Python

Common Use Cases

• Parsing complex, layout-heavy PDFs (tables, forms, scanned documents) into clean markdown for downstream LLM consumption
• Extracting structured JSON (invoices, contracts, forms) from unstructured documents for automated data-entry pipelines
• Standing up a managed RAG retrieval index without operating your own vector store or ingestion pipeline
• Adding a document-parsing preprocessing step to an existing LlamaIndex or LangChain RAG pipeline

Under The Hood

Architecture: The package (py/llama_cloud_services/) is organized as one subpackage per service — parse/ (with its own cli/ and types.py), extract/ (extract.py, utils.py), and index/ — each wrapping HTTP calls to the corresponding LlamaCloud API endpoint behind a typed client class, with constants.py centralizing default base URLs (including the EU endpoint) and a beta/ package for services still under active iteration.

Tech Stack: Pure Python, built with hatchling as the build backend and uv for dependency locking (uv.lock). The repo is a pnpm-workspace monorepo that also ships a parallel TypeScript client (ts/) alongside the Python package, sharing docs and examples across both language SDKs.

Code Quality: 24 test files across tests/ and unit_tests/, plus mypy type checking and pre-commit hooks configured in the dev dependency group, indicating a maintained CI/quality bar even as the repo transitions toward deprecation. The MIT license and clear per-service module boundaries keep the codebase easy to navigate.

API Design: Getting started is a single pip install llama-cloud-services plus one API key shared across all three service clients, with a consistent constructor pattern (api_key=, optional base_url=) across LlamaParse, LlamaExtract, and LlamaCloudIndex. Documentation is split across per-service quickstart guides rather than one combined reference, and the package now carries a prominent deprecation notice directing new users to the successor llama-cloud package, which is worth factoring into any new adoption decision.