LangChain
LangChain (Python)
LangChain's Python ecosystem continued steady incremental progress across core and partner packages. The core library shipped security fixes for prompt loading and symlink handling, while partner integrations expanded model support: Ollama added response format and logprobs support, Anthropic introduced prompt caching middleware and automatic compaction for Opus 4.6, and OpenAI shipped tool search and server-side compaction. Across the board, model profile data refreshed regularly and token counting improved to account for tool schemas and context overflow scenarios. The main package optimized agent creation speed by 15% and added LangSmith tracing metadata to create_agent and init_chat_model flows, while text-splitters and standard test suites received incremental refinements for mutation prevention and sandbox provider coverage.
March focused on stability and model integration across the ecosystem. The main package shipped performance improvements—a 15% reduction in init speed for create_agent—alongside security patches for Pygments and dependency bumps across cryptography and requests. Model profile infrastructure expanded significantly through langchain-openrouter and langchain-anthropic, with OpenRouter gaining streaming token usage support and Anthropic's prompt caching middleware now applying explicit caching to system messages and tool definitions. A ReDoS vulnerability in MRKL and ReAct action parsing was backported to v0.3, and langchain-core added anti-SSRF hardening while BaseCrossEncoder moved into core for broader access.