Skip to content

LangChain Adapter

Monitor tool calls in LangChain agents via middleware callback.

Install

pip install toolwitness[langchain]

Usage

from toolwitness.adapters.langchain import ToolWitnessMiddleware
from toolwitness.storage.sqlite import SQLiteStorage

middleware = ToolWitnessMiddleware(
    on_fabrication="raise",  # or "log" or "callback"
    storage=SQLiteStorage(),
)

# Add as a callback to your LangChain agent
agent = create_agent(llm, tools, callbacks=[middleware])

Fabrication handling modes

Mode Behavior
"log" Log the failure, continue execution
"raise" Raise a FabricationDetectedError
"callback" Call a custom function with the verification result

Custom callback

def on_failure(result):
    print(f"Failure detected: {result.tool_name} — {result.classification}")

middleware = ToolWitnessMiddleware(
    on_fabrication="callback",
    fabrication_callback=on_failure,
    storage=SQLiteStorage(),
)

Options

Parameter Type Default Description
on_fabrication str "log" How to handle detected fabrications
fabrication_callback callable None Custom handler (when mode is "callback")
storage SQLiteStorage None Persist results for CLI and dashboard
session_id str auto-generated Custom session identifier

How it integrates

The middleware hooks into LangChain's callback system:

  • on_tool_start — records the tool call and generates a receipt
  • on_tool_end — records the tool output
  • on_llm_end — verifies the model's response against recorded outputs

This means verification happens automatically as part of the agent's execution loop.

Next