auth0-ai-llamaindex 0.1.1

This package is an SDK for building secure AI-powered applications using Auth0, Okta FGA and LlamaIndex.

Auth0 AI for LlamaIndex

auth0-ai-llamaindex is an SDK for building secure AI-powered applications using Auth0, Okta FGA and LlamaIndex.

Installation

[!WARNING] auth0-ai-llamaindex is currently under development and it is not intended to be used in production, and therefore has no official support.

pip install auth0-ai-llamaindex

Async User Confirmation

Auth0AI uses CIBA (Client Initiated Backchannel Authentication) to handle user confirmation asynchronously. This is useful when you need to confirm a user action before proceeding with a tool execution.

Full Example of Async User Confirmation.

Define a tool with the proper authorizer specifying a function to resolve the user id:

from auth0_ai_llamaindex.auth0_ai import Auth0AI
from auth0_ai_llamaindex.ciba import get_access_token
from llama_index.core.tools import FunctionTool

auth0_ai = Auth0AI()
with_async_user_confirmation = auth0_ai.with_async_user_confirmation(
    scope="stock:trade",
    audience=os.getenv("AUDIENCE"),
    user_id=lambda _ctx: session["user"]["userinfo"]["sub"]
    binding_message=lambda ctx: f"Authorize the purchase of {ctx['qty']} {ctx['ticker']}",
)

def tool_function(ticker: str, qty: int) -> str:
    access_token = get_access_token()
    headers = {
        "Authorization": f"{access_token["type"]} {access_token["value"]}",
        # ...
    }
    # Call API

trade_tool = with_async_user_confirmation(
    FunctionTool.from_defaults(
        name="trade_tool",
        description="Use this function to trade a stock",
        fn=tool_function,
        # ...
    )
)

Authorization for Tools

The FGAAuthorizer can leverage Okta FGA to authorize tools executions. The FGAAuthorizer.create function can be used to create an authorizer that checks permissions before executing the tool.

Full Example of Authorization for Tools.

1. Create an instance of FGA Authorizer:

from auth0_ai_langchain.fga.fga_authorizer import FGAAuthorizer, FGAAuthorizerOptions

fga = FGAAuthorizer.create()

Note: Here, you can configure and specify your FGA credentials. By default, they are read from environment variables:

FGA_STORE_ID="<fga-store-id>"
FGA_CLIENT_ID="<fga-client-id>"
FGA_CLIENT_SECRET="<fga-client-secret>"

2. Define the FGA query (build_query) and, optionally, the on_unauthorized handler:

def build_fga_query(tool_input):
    return {
        "user": f"user:{context.get("user_id")}",
        "object": f"asset:{tool_input["ticker"]}",
        "relation": "can_buy",
        "context": {"current_time": datetime.now(timezone.utc).isoformat()}
    }

def on_unauthorized(tool_input):
    return f"The user is not allowed to buy {tool_input["qty"]} shares of {tool_input["ticker"]}."

use_fga = fga(FGAAuthorizerOptions(
    build_query=build_fga_query,
    on_unauthorized=on_unauthorized,
))

Note: The parameters given to the build_query and on_unauthorized functions are the same as those provided to the tool function.

3. Wrap the tool:

from llama_index.core.tools import FunctionTool

async def buy_tool_function(ticker: str, qty: int) -> str:
        # TODO: implement buy operation
        return f"Purchased {qty} shares of {ticker}"

func=use_fga(buy_tool_function)

return FunctionTool.from_defaults(
    fn=func,
    async_fn=func,
    name="buy",
    description="Use this function to buy stocks",
)

Calling APIs On User's Behalf

The Auth0AI.with_federated_connection function exchanges user's refresh token for a Federated Connection API access token.

Full Example of Calling APIs On User's Behalf.

Define a tool with the proper authorizer specifying a function to resolve the user's refresh token:

from auth0_ai_llamaindex.auth0_ai import Auth0AI
from auth0_ai_llamaindex.federated_connections import get_access_token_for_connection
from llama_index.core.tools import FunctionTool

auth0_ai = Auth0AI()

with_google_calendar_access = auth0_ai.with_federated_connection(
    connection="google-oauth2",
    scopes=["https://www.googleapis.com/auth/calendar.freebusy"],
    refresh_token=lambda *_args, **_kwargs: session["user"]["refresh_token"],
)

def tool_function(date: datetime):
    access_token = get_access_token_for_connection()
    # Call Google API

check_calendar_tool = with_google_calendar_access(
    FunctionTool.from_defaults(
        name="check_user_calendar",
        description="Use this function to check if the user is available on a certain date and time",
        fn=tool_function,
        # ...
    )
)

RAG with FGA

The FGARetriever can be used to filter documents based on access control checks defined in Okta FGA. This retriever performs batch checks on retrieved documents, returning only the ones that pass the specified access criteria.

Full Example of RAG Application.

from llama_index.core import VectorStoreIndex, Document
from auth0_ai_llamaindex import FGARetriever
from openfga_sdk.client.models import ClientCheckRequest
from openfga_sdk import ClientConfiguration
from openfga_sdk.credentials import CredentialConfiguration, Credentials

# Define some docs:
documents = [
    Document(text="This is a public doc", doc_id="public-doc"),
    Document(text="This is a private doc", doc_id="private-doc"),
]

# Create a vector store:
vector_store = VectorStoreIndex.from_documents(documents)

# Create a retriever:
base_retriever = vector_store.as_retriever()

# Create the FGA retriever wrapper:
retriever = FGARetriever(
    base_retriever,
    build_query=lambda node: ClientCheckRequest(
        user=f'user:{user}',
        object=f'doc:{node.ref_doc_id}',
        relation="viewer",
    )
)

# Create a query engine:
query_engine = RetrieverQueryEngine.from_args(
    retriever=retriever,
    llm=OpenAI()
)

# Query:
response = query_engine.query("What is the forecast for ZEKO?")

print(response)

---

Auth0 is an easy to implement, adaptable authentication and authorization platform. To learn more checkout Why Auth0?

This project is licensed under the Apache 2.0 license. See the LICENSE file for more info.