v0.14.3

Breaking Changes

📅 Sep 24, 2025📦 llamaindexView on GitHub →

⚠ 2 breaking✨ 7 features🐛 4 fixes🔧 11 symbols

Summary

This release introduces ThinkingBlock support across major LLM providers, adds new PaddleOCR and Azure PostgreSQL integrations, and migrates MongoDB components to the PyMongo Asynchronous API. It also includes a critical security fix for SingleStoreDB and updates the Firecrawl reader to the v2 SDK.

⚠️ Breaking Changes

The MongoDB-related packages (readers, chat-store, kvstore) have migrated from the Motor library to the PyMongo Asynchronous API. This may break custom implementations relying on Motor-specific objects.
The Firecrawl web reader has been migrated to the Firecrawl v2 SDK, which may involve breaking changes in configuration or response schemas.

Migration Steps

Update MongoDB-based integrations to ensure compatibility with the PyMongo Asynchronous API instead of Motor.
Upgrade Firecrawl SDK to v2 when using llama-index-readers-web.

✨ New Features

Introduced ThinkingBlock to content blocks across core and multiple LLM providers (Anthropic, Google GenAI, MistralAI, OpenAI).
Added llama-index-readers-paddle-ocr: A new package for extracting text from images in PDFs using PaddleOCR.
Added llama-index-vector-stores-azurepostgresql: New support for Azure PostgreSQL vector storage.
Added FileAPI support for documents in Google GenAI LLM.
Added kimik2 0905 support and dynamic model API validation for Baseten.
Added Valyu Extractor and Fast mode to Valyu tools.
Improved state snapshotting in AG-UI protocols.

🐛 Bug Fixes

Fixed Gemini thought signature serialization in llama-index-core.
Fixed a SQL injection vulnerability in SingleStoreDB vector store.
Fixed compatibility issues for old sparse vectors in Qdrant.
General fixes for Azure AI Search vector store.

🔧 Affected Symbols

ThinkingBlockGeminiFirecrawlReaderPaddleOCRReaderAzurePostgreSQLVectorStoreMongoReaderMongoChatStoreMongoKVStoreValyuExtractorQdrantVectorStoreSingleStoreVectorStore