langchain langchain-core==1.7.0
- Source
- github.com
- Author
- github-actions[bot]
- Date

Context
When a chat model decides to call a tool, it generates the call's arguments piece by piece, and LangChain streams them to your code as they arrive. The langchain-core 1.7.0 release notes say this release changes the shape of that stream. Arguments used to arrive as block-delta events, each a snapshot of all the arguments produced so far. They now arrive as tool-call-delta events, each holding only the new slice, so code that wants the full arguments mid-stream has to join the slices itself.
The notes spell out who is affected. Code that reads raw stream events and looks at block-delta arguments partway through should read tool-call-delta instead. The higher-level ways of getting results, .tool_calls.get() and .output, are unchanged, so code that waits for the finished call shouldn't notice. Anyone who needs the old snapshots can pin langchain-core below 1.7. The notes add that since 1.6.13, looping over ChatModelStream.tool_calls already yields each call's new argument slice, not the running total.
The release itself contains one feature, PR 41257, which emits tool-call-delta for streamed tool-call arguments, plus the version bump from 1.6.13. langchain-core is the base package that LangChain's other libraries build on, so this change reaches anything that reads its chat-model stream events directly.
- streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
Checking sign-in…
Loading comments…





