Repository Issue Activity (beta)

#39881 create_agent: heterogeneous middleware list collapses inferred state to AgentState, rejecting middleware with richer state schemas

#39880 core: `BaseMessageChunk.__add__` rejects tuple of chunks while `AIMessageChunk` accepts it

#39842 PIIMiddleware `ip` type does not detect IPv6 addresses

#38209 Silent quality degradation in multi-agent flat round-table: 34.3/100 without synthesis node vs 86.3/100 with it

#39835 core: `DictPromptTemplate.aformat` is never called — the async prompt path uses sync `format`

#39856 Raises: block on Runnable.__ror__ states a rationale that only holds for __or__

#39858 count_tokens_approximately counts every image as 85 tokens regardless of resolution

#39870 Add a first-party Python DirectoryLoader with a per-extension loader map (JS parity)

#38892 RunnableWithFallbacks.stream()/astream() misinterpret a legitimately empty upstream stream as a failure

#39877 feat(langchain): wrap GraphBubbleUp with agent context in middleware

#39876 Feature request: HOL Guard tool-call middleware integration for LangChain agents

#37248 configurable field lost in ToolRuntime.config when using tool_interceptors in langchain 1.2.16 (works in 1.2.15)

#39874 _rm_titles does not recurse into lists, so pydantic titles survive inside anyOf branches of tool schemas

#39820 `StateFileSearchMiddleware`: `grep_search` and `glob_search` return no results when `path` ends in a slash

#39871 ollama: ChatOllama drops AIMessage.invalid_tool_calls when converting history

#39743 Prompt templates silently misrender or reject non-`dict` `Mapping` inputs (`ChainMap`, `MappingProxyType`, `UserDict`)

#39713 `ChatAnthropic(stream_usage=False)` also drops `stop_reason`, `stop_sequence` and `model_name` from `response_metadata`, not just usage metadata

#39857 count_tokens_approximately counts every image as 85 tokens regardless of resolution

#39747 Streamed refusals are dropped, so structured output raises a generic `ValueError` instead of `OpenAIRefusalError`

#32120 Reasoning model with structured output

#31647 Add Perplexity cost tracking

#28508 The batch method from ChatModels and all the Runnables does not really support the OpenAI batch API.

#32197 Support structured output with Pydantic models in `langchain-huggingface`

#38223 Responses API with streaming and a nested Pydantic schema results in invalid strict JSON schema (additionalProperties/required missing on $defs)

#39776 Add langchain-opensearch as a standalone integration package (follow-up to #30323)