Access control for AI agents (RBAC) using Airtable and Telegram

Purpose

This workflow allows granular control over the access to tools connected to AI Agents (including Multi-Agent setups) using Role Based Access Control.

Demo & Explanation

How it works

• User permissions are managed in Airtable where every restricted AI tool is listed by name and connected via roles to users
• Requests to the Main Agent can be sent through a Telegram message (can be replaced by Whatsapp, IMAP or similar)
• On every request the Telegram username is used to query a list of all allowed tools which are linked in Airtable
• A LangChain Code node is used to compare that list against the connected tools
• Every tool which is not permitted to be used is being replaced by a tool, which has a status response, telling the Agent to return a message to the user, that he is not authorized to use the tool
• Otherwise allowed tools are passed through to the Agent, as if they were connected directly to the Agent
• The parameters can also be passed to a sub-agent called as a sub-workflow where permissions can be checked the same way
• Every response is sent back to the same Telegram conversation

Setup

• Clone the workflow and select the belonging credentials. You'll need an OpenAI and Airtable Account as well as a Telegram Bot (refer to the docs for the Telegram credentials).
• Copy this Airtable Template into your workspace
• Follow the instructions given in the yellow sticky notes
• Activate the workflow

How to use

Try this example:

• Create a new line in Airtable under “Users” containing your Telegram username and your full name
• Set the roles “basic” and “info”
• Consider temporarily disconnecting or resetting the chat memories so they do not remember previous confirmations
• Start a new chat, asking about your permitted roles - you should get a list of those
• Ask about the current weather in your city - you should be informed, that you do not have permission to access that information
• Back in Airtable add the role “weather” to your user
• Now ask the Agent the same question again - It should give you a proper answer this time

From here on you can add tools and create roles to your likings.

Disclaimer

Please note, that this workflow can only run on self-hosted n8n instances, since it requires the LangChain Code Node.