pypi / langchain-classic
langchain-classic
What changed in langchain-classic 1, from its official release notes, upgrade guides and advisories. Sources checked 2026-11-09.
4 breaking4 security4 deprecations4 new APIs and options3 behavior and defaults1 tooling6 fixes and other
See which of these apply to your code, ranked, in your agent:
npx @unstale/cli initlangchain-classic 126 changes
Breaking changes 4
Minimum core versions were bumped, so projects may need to use a newer langchain-core version.
BaseCrossEncoder moved to langchain-core, changing where developers obtain this API.
The prefix keyword is now keyword-only to match BaseStore.
Tigris is removed.
Security 4
LangSmith SDK: Public prompt pull deserializes untrusted manifests without trust boundary warning
Deserialization is restricted in this storage module.
The load() function is hardened against untrusted manifests.
A ReDoS vulnerability in the MRKL and ReAct action regex was patched.
Deprecations 4
The hub API is deprecated.
Loads and dumps are limited; the notes do not specify the limits.
Deprecation targets now point to create_agent, changing the suggested replacement for deprecated APIs.
Deprecations were added; the notes do not specify which APIs are affected.
New APIs and options 4
OpenAI support was added for automatic server-side compaction.
BaseTool gains extras support.
Anthropic support adds the effort option.
The google_genai provider is now supported in init_embeddings.
Behavior and default changes 3
Supports state updates from wrap_model_call with command(s).
Improved validation changes how init_chat_model handles its inputs.
Run IDs now use uuid7.
Tooling 1
Typing for Runnable.__or__ is improved.
Fixes and other changes 6
Callbacks now support content block tokens.
The deprecation for hub.pull uses the langchain-classic version.
Model provider classes were updated for Azure AI Foundry.
Weights are now validated, helping catch invalid ensemble retriever configurations.
Provider prefixes are decoupled from model names when initializing chat models.
The HuggingFace backend helper logic for init_chat_model is fixed.