LangChain
The LangChain integration wraps a compatible executor or Runnable and attaches Arclasp governance callbacks around supported execution events. Use the LangGraph integration instead when your application is primarily a compiled LangGraph workflow.Install
langchain-core>=0.3.0,<0.4.0.
Your application may need additional LangChain packages and provider integrations. For example, code that imports langchain.agents or langchain_openai must install those packages separately.
Add Arclasp to an existing executor
This is an integration fragment. It assumesexecutor already exists in your application.
What the adapter records
Where the underlying Runnable emits compatible callbacks, Arclasp can record events around supported tool, LLM, and chain lifecycle activity. Tool-call governance is the important enforcement boundary: a governed tool invocation is evaluated before the tool body is allowed to proceed. If policy denies the governed tool call, Arclasp raises and the tool should not execute through that governed path. If approval is required, the governed call waits for the approval result.Existing callbacks
Arclasp appends its callback behavior to compatible existing callback configuration rather than requiring you to remove your own callbacks.Agent name
The current wrapper derives the recorded agent label from the wrapped object’s class name. If you need precise agent naming for evidence, instrument the workflow manually witharclasp.Chain.record_agent_action().