LlamaIndex

The SDK ships LlamaIndex adapters for both languages. The wrapped tool keeps its name, description, and schema, so it is a drop-in replacement.

npm install @axiorank/sdk

guardTools wraps a tools array so each call's arguments are scored before the tool runs. The adapter is structural (anything with call + metadata qualifies), so it works with FunctionTool, tool(), and custom BaseTool implementations.

import { AxioRank } from "@axiorank/sdk";
import { guardTools } from "@axiorank/sdk/llamaindex";
import { agent } from "@llamaindex/workflow";

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

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

Recover or fail

onDeny: "return" returns a short, model-readable refusal the agent can re-plan around. The default throws AxioRankDeniedError instead, failing the step hard.

pip install "axiorank[llamaindex]"

guard_tool wraps a LlamaIndex FunctionTool. Provide client (sync), async_client (async), or both; LlamaIndex derives the direction it needs from whichever you supply.

from axiorank import AxioRank
from axiorank.integrations.llamaindex import guard_tool
from llama_index.core.tools import FunctionTool

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

tool = FunctionTool.from_defaults(fn=deploy_service)
safe_tool = guard_tool(tool, axio, on_deny="return")
# ...hand safe_tool to your agent as usual.

Recover or fail

on_deny="return" returns a short, model-readable refusal the agent can re-plan around. The default, on_deny="raise", raises AxioRankDeniedError instead.

Next steps

Framework integrations: the shared model and axio.trace().
Python SDK · TypeScript SDK: the underlying clients.

LlamaIndex

Next steps

On this page