Class●Since v3.0

ModelArmorMiddleware

Middleware to integrate Model Armor sanitization into agent execution.

This middleware provides hooks that sanitize user prompts before they reach the model and sanitize model responses before they're returned to the user.

Sanitization is enabled by providing the corresponding runnable:

Provide prompt_sanitizer to enable user prompt sanitization
Provide response_sanitizer to enable model response sanitization

ModelArmorMiddleware(
  self,
  *,
  prompt_sanitizer: ModelArmorSanitizePromptRunnable | None = None,
  response_sanitizer: ModelArmorSanitizeResponseRunnable | None = None
)

Bases

lc_agents_middleware.AgentMiddleware

Example:

from langchain.agents import create_agent
from langchain_google_vertexai import ChatVertexAI
from langchain_google_community.model_armor import (
    ModelArmorMiddleware,
    ModelArmorSanitizePromptRunnable,
    ModelArmorSanitizeResponseRunnable,
)

# Create sanitizer runnables
prompt_sanitizer = ModelArmorSanitizePromptRunnable(
    project="my-project",
    location="us-central1",
    template_id="my-template",
    fail_open=False,
)
response_sanitizer = ModelArmorSanitizeResponseRunnable(
    project="my-project",
    location="us-central1",
    template_id="my-template",
    fail_open=False,
)

# Create middleware with both sanitizers
middleware = ModelArmorMiddleware(
    prompt_sanitizer=prompt_sanitizer,
    response_sanitizer=response_sanitizer,
)

# Create agent with Model Armor protection
agent = create_agent(
    model=ChatVertexAI(model_name="gemini-2.0-flash-001"),
    tools=[...],
    middleware=[middleware],
)

# Or create middleware with only prompt sanitization
middleware = ModelArmorMiddleware(prompt_sanitizer=prompt_sanitizer)

Parameters

Name	Type	Description
`prompt_sanitizer`	`ModelArmorSanitizePromptRunnable \| None`	Default:`None` Runnable for sanitizing user prompts before model calls. If provided, prompt sanitization is enabled. If None, prompts are not sanitized.
`response_sanitizer`	`ModelArmorSanitizeResponseRunnable \| None`	Default:`None` Runnable for sanitizing model responses. If provided, response sanitization is enabled. If None, responses are not sanitized.