🚀 New Features

• Context Editing Middleware (#9118): Added middleware for editing conversation context dynamically
• Tool Call Limit Middleware (#9121): Implemented middleware to limit the number of tool calls during agent execution
• Retry Model Request Hook (#9128): Added retry functionality for model requests in middleware
• Call Count Tracking (#9090): Introduced call counts through private state for better monitoring and control
• Custom Interrupted Tool Descriptions (#9116): Added ability to pass in a description factory for interrupted tool calls
• PII Redaction Middleware (#9058): Added middleware for automatically redacting personally identifiable information
• Perplexity Support: Added Perplexity provider support to universal chat model

🐛 Bug Fixes

• LLM Tool Selector Implementation (#9127): Updated implementation of LLMToolSelector middleware for improved reliability
• LangGraph Integration (#9117): Fixed integration issues with LangGraph
• Tool Instance Handling (#9119): Fixed tools in ModelRequest to return tool instances instead of serialized objects
• TypeScript Type Inference (#9087): Exported all symbols to enable proper TypeScript type inference
• Prompt Caching Immutability (#9086): Fixed PromptCachingMiddleware to return new message objects instead of modifying existing ones
• Input Type Support (#9080): Improved input types to support a wider range of type definitions
• Tools Node Creation (#9074): Fixed to only create tools node for client-provided tools
• Planning Middleware Adjustments (#9070): Minor improvements to planning middleware behavior
• Hub Prompt Schema Binding (#9057): Fixed schema binding for various types of pulled hub prompts

🚀 New Features

• Todo Middleware (#9051): Added new todo middleware functionality for agent workflows
• LLM Tool Selector Middleware (#9050): Implemented llmToolSelectorMiddleware (bigTool) for intelligent tool selection
• Middleware Tool Definition (#9046): Middleware can now define their own tools dynamically
• Dynamic Prompt Middleware (#8993): Added built-in middleware for dynamic prompt generation
• Unsupported Model Behavior (#9039): Added unsupportedModelBehavior property to anthropic middleware

🐛 Bug Fixes

• System Prompt Consistency (#9045): Fixed system prompt handling across all agent components
• Zod Interoperability (#9029): Improved compatibility between Zod v3 and v4
• Structured Output Handling (#9033): Improved structured output processing for agents
• Graph Support for Jumps (#9035): Added statically typed graph support for agent jumps
• Tool Configuration Naming (#9034): Renamed toolConfigs to interruptOn for clarity
• LLM Property Removal (#9017): Removed llm prop from middleware agent
• Message Handling: Fixed grabbing last AIMessage from list instead of just looking at last message
• Memory Saver Export (#8996): Removed incorrect MemorySaver re-export
• Middleware Naming (#8912): Fixed "middlewares" property naming to "middleware"
• Cache Control Verification: Actually verify cache control values in usage response
• Message Instance Checking (#8905): Use isInstance for message validation

📦 Dependency and Build Fixes

• OpenAI Provider Dependency (#8920): Removed dependency to openai provider package (alpha.6)
• Internal ESLint Package (#8770): Created internal ESLint package (alpha.6)
• Zod v4 Support (#8792): Updated to support Zod v4 (alpha.6)
• Test Console Logs (#8998): Removed console logs in test files (alpha.7)

🚀 New Features

• Google AI Standard Content (#9024): Added comprehensive standard content support for Google GenAI and Vertex AI
• AWS Bedrock Standard Content (#9013): Implemented standard content support for AWS Bedrock Converse
• Server Tool Blocks (#9000): Added server tool blocks to standard content system
• Zod Interoperability (#9029): Improved compatibility between Zod v3 and v4

🐛 Bug Fixes

• Response Metadata Type Mismatch (#9010): Fixed type mismatches in response metadata
• Message Index Cleanup (#8972): Refined message exports

🔧 Technical Improvements

• Google AI Integration: Significantly enhanced Google GenAI and Vertex AI block translators
• AWS Bedrock Integration: Major improvements to Bedrock Converse support
• Content Index Updates: Enhanced content module exports

What's Changed

• feat(langchain) new middleware concept for createAgent by @christian-bromann in #8851

Full Changelog: langchain==1.0.0-alpha.4...langchain==1.0.0-alpha.5

What's Changed

• fix(core): Fix deep nesting of runnables within traceables by @jacoblee93 in #8853

Full Changelog: https://github.com/langchain-ai/langchainjs/compare/@langchain/core==0.3.74...@langchain/core==0.3.75

What's Changed

• fix(community): bump firebase-admin devDependencies by @hntrl in #8861
• fix(community): URL encode paths in GitHub document loader by @hntrl in #8860

Full Changelog: https://github.com/langchain-ai/langchainjs/compare/@langchain/community==0.3.54...@langchain/community==0.3.55

What's Changed

• fix(anthropic): opus-4.1 thinking by @taiichiro-nagase in #8856

New Contributors

• @taiichiro-nagase made their first contribution in #8856

Full Changelog: https://github.com/langchain-ai/langchainjs/compare/langchain==0.3.33...@langchain/anthropic==0.3.27

What's Changed

• fix(langchain): handle prompts created from runs by @hntrl in #8848

Full Changelog: langchain==0.3.32...langchain==0.3.33

What's Changed

• fix(core): Always inherit parent run id onto callback manager from context by @jacoblee93 in #8849

Full Changelog: https://github.com/langchain-ai/langchainjs/compare/@langchain/core==0.3.73...@langchain/core==0.3.74

What's Changed

• fix(langchain): add support for conversion of ref in array schema by @jnjacobson in #8719

New Contributors

• @jnjacobson made their first contribution in #8719

Full Changelog: langchain==0.3.31...langchain==0.3.32