LangChain 1.4.2 fixes tool calls in HITL edits
The patch release logs two changes since 1.4.1: a fix that keeps model-generated tool calls intact when a human edits them, plus a ToolMessage notice.
In short
LangChain 1.4.2 is a two-entry patch release whose only functional change, pull request #40463, preserves model-generated tool calls when a person edits them in a human-in-the-loop step and adds a notice to the ToolMessage.
At a glance
- Version 1.4.2 of the langchain package; the release page dates it September 18.
- Changelog against 1.4.1: exactly two entries — release commit #40621 and bug fix #40463.
- The fix keeps model-generated tool calls in place when a human edits the tool call in the HITL step.
- It also adds a notice to the ToolMessage; the page does not spell out that notice's wording.
- The page points at commit a18de590e7ccf5c647fbf3d689e5f1a15f78e9f5.
LangChain 1.4.2 ships a single functional change: pull request #40463 stops model-generated tool calls from being dropped when a person edits them in a human-in-the-loop step, and attaches a notice to the ToolMessage.
The whole changelog is two lines
Measured against 1.4.1, the release page shows two entries — the version bump itself, #40621, and the fix. The page pins the release to commit a18de590e7ccf5c647fbf3d689e5f1a15f78e9f5 and dates it September 18.
Why an edited tool call can go missing
A human-in-the-loop agent pauses before execution so a reviewer can approve the proposed call or rewrite its arguments. The fix makes the model's original tool calls survive that edit rather than disappear with it. What the notes leave out is the failure path itself: which call shapes were affected, and since which version.
Read the notes, not the version number
When the page was retrieved, the compare view and the asset list did not fully render. Contributors, the complete commit range and any breaking-change statement therefore cannot be sourced from it. Semantic versioning implies a patch carries no break, but the page never says so in words.
Who should take the upgrade
Teams running an approval gate in front of tool execution get the practical benefit, because a truncated message history is what makes an agent lose the thread on its next turn. Everyone else is looking at a version bump with no second behavior change attached. Both entries are listed in full in the release notes.
FAQ
What changed in LangChain 1.4.2?
Against 1.4.1 the release page lists two entries: release commit #40621 and bug fix #40463, which covers tool calls during HITL edits.
What does the HITL tool call fix actually do?
When a human edits a proposed tool call, the model-generated tool calls are preserved instead of dropped, and the ToolMessage carries a notice.
Should I upgrade from LangChain 1.4.1?
If your agents pause for human approval before running tools, 1.4.2 is the only entry that changes behavior. The page makes no breaking-change statement either way.
