LangChain · LangGraph

npm install @axiorank/sdk @langchain/core

guardTools wraps a tools array so each call's arguments are scored before the tool runs. The guarded tool keeps the original name, description, and schema, making it a drop-in anywhere the original is used (LangChain or LangGraph).

import { AxioRank } from "@axiorank/sdk";
import { guardTools } from "@axiorank/sdk/langchain";
import { createReactAgent } from "@langchain/langgraph/prebuilt";

const axio = new AxioRank({ apiKey: process.env.AXIORANK_API_KEY! });

const agent = createReactAgent({
  llm,
  tools: guardTools(myTools, axio, { onDeny: "return" }),
});

Recommended: one middleware for the whole agent. AxioRankMiddleware scores every tool call before it runs and every model turn as it happens, and it plugs straight into create_agent. A denied tool call comes back as a model-readable refusal so the agent re-plans; a redact policy masks secrets and PII in the model's output.

pip install langchain-axiorank

from langchain.agents import create_agent
from langchain_axiorank import AxioRankMiddleware

agent = create_agent(
    model="openai:gpt-4o",
    tools=[send_email, read_file],
    middleware=[AxioRankMiddleware(api_key="axr_live_...")],
)

Set AXIORANK_API_KEY and you can drop the api_key argument. Building from a key wires up both a sync and an async client, so the same middleware works with agent.invoke and agent.ainvoke. Pass an axio.trace() handle as client to correlate the run on the Agent Runs page.

Existing agents: a callback handler. On a classic AgentExecutor, attach AxioRankCallbackHandler (from axiorank[langchain]) and every tool the agent runs is checked first. A denied call raises AxioRankDeniedError, aborting the step.

pip install "axiorank[langchain]"

from axiorank import AxioRank
from axiorank.integrations.langchain import AxioRankCallbackHandler

axio = AxioRank(api_key="axr_live_...")

agent_executor.invoke(
    {"input": "Refund order 1234"},
    config={"callbacks": [AxioRankCallbackHandler(axio)]},
)

For async agents, use AxioRankAsyncCallbackHandler with an AsyncAxioRank client. A callback can allow or stop a call but cannot substitute a tool's output, so a blocked call raises. To let the agent recover from a denial, use the middleware above, or wrap a single tool with guard_tool (on_deny="return"):

from axiorank.integrations.langchain import guard_tool

safe_tool = guard_tool(my_tool, axio, on_deny="return")