PageIndex v0.2.20 Ships SDK, Flash Engine
PageIndex v0.2.20 delivers an end-to-end Python SDK for vectorless, reasoning-based RAG, with local and cloud clients covering document submission, chat, and agent integration. Its Flash engine derives document trees from layout statistics, parallelizes expansion, and runs locally without a server, vector database, or PageIndex API key.
This release makes PageIndex substantially more practical for developers who want structured document retrieval without operating a vector stack, though local mode remains focused on text-based PDFs.
- –Flash removes LLMs from structure generation, reserving them for summaries and reasoning.
- –A shared local/cloud API reduces migration friction from prototypes to hosted deployments.
- –Bring-your-own-model support separates document hosting from answer generation.
- –Tree navigation offers a stronger fit for cross-references, long reports, and traceable answers than similarity-only retrieval.
- –Local mode trades convenience for narrower support: cloud handles scanned and image-rich documents, while local indexing targets text-based PDFs.
DISCOVERED
2h ago
2026-09-29
PUBLISHED
1d ago
2026-09-28
RELEVANCE