Llamaparse

The LlamaParse API follows a submit-and-poll pattern: POST a file to /upload, receive a job_id, GET /results/{job_id} until complete. Output format is configurable (markdown, json, text). The API accepts multipart/form-data uploads and returns JSON. Authentication is via api-key header. The surface is small and well-defined — upload, poll, retrieve. LlamaIndex Python SDK wraps this cleanly. Rate limits and concurrent job limits are documented in the free and paid tiers.

LlamaParse is async by design — parsing large documents takes seconds to minutes. The poll-until-complete pattern means agents must handle timeouts gracefully. Failed jobs return an error status in the results endpoint with a message field. HTTP errors (4xx for bad input, 5xx for service issues) are standard JSON. No retry-after headers documented for rate limits. The service has documented SLAs for the paid tier.

LlamaParse is a cloud document parsing service purpose-built for AI and RAG pipelines. It accepts PDFs, DOCX, PPTX, and images via a REST API, and returns clean markdown, structured JSON, or page-level text with table and image extraction. Agents can upload documents, poll for job completion, and retrieve parsed results — the API is designed for async processing of large documents. A free tier offers a limited number of pages per day. Integration with LlamaIndex and LangChain is first-class, but the REST API is fully usable standalone. Confidence is docs-derived; latency and quality of complex table extraction are not runtime-verified.

Authentication uses an API key passed as the api-key header or via the LlamaCloud dashboard. Keys are scoped per project. No OAuth or per-request signing. Free tier keys are rate-limited. Revocation is through the dashboard. No IP allowlisting documented. Standard API key security model — straightforward for agent use.

Documentation is hosted at docs.cloud.llamaindex.ai. Covers API reference, authentication, output formats, and integration guides for LlamaIndex and LangChain. Examples are Python-first. REST API docs are present but less detailed than SDK docs. LlamaCloud dashboard provides usage metrics. Community support via LlamaIndex Discord. Quality is good for the primary use case; edge cases in table/image extraction are less documented.