{"id":411,"date":"2025-09-19T13:40:34","date_gmt":"2025-09-19T12:40:34","guid":{"rendered":"https:\/\/guillaumesblog.net\/?p=411"},"modified":"2025-09-19T14:48:21","modified_gmt":"2025-09-19T13:48:21","slug":"rapid-set-up-of-ai-agents-in-the-cloud","status":"publish","type":"post","link":"https:\/\/guillaumesblog.net\/index.php\/rapid-set-up-of-ai-agents-in-the-cloud\/","title":{"rendered":"Hook Your Tools Into Cloud AI Agents \u2014 Fast"},"content":{"rendered":"\n

Everyone\u2019s talking about \u201cAgentic AI,\u201d so I decided to experiment. My goal: a custom agent that can leverage my data and tools, set up quickly in the cloud using OCI.<\/p>\n\n\n\n\n\n\n\n

Let’s imagine we want it to be a boss at calculus and we don’t want the agent and its embedded LLM to do it – because we know this is something they are (were?) bad at – We want to subcontract the “maths” to a “tool” so the LLM could decide to call on it autonomously if required.<\/p>\n\n\n\n

Disclaimer: This guide is for demonstration only, I assume you have a good comprehension of the cloud & its setup, you know there is a cost involved and that I am not securing the APIs. You should always secure API in a production setting.<\/em> I also trust that you know what you are doing about firewalls, VCN, subnets and the like. All of it is not the point here so I won’t address it.<\/em><\/p>\n\n\n\n

Here is what we are going to do:<\/p>\n\n\n\n

    \n
  1. Create a simple back-end with 4 random maths functions and APIs.<\/li>\n\n\n\n
  2. Launch the back-end.<\/li>\n\n\n\n
  3. Create an agent on my cloud console.<\/li>\n\n\n\n
  4. Register a tool now part of the agent capabilities.<\/li>\n\n\n\n
  5. Prompt and verify function calling by the agent.<\/li>\n<\/ol>\n\n\n\n

    Let’s go – Get your VM up, python, pip, venv, copy paste the code<\/a> available in my repo and save the file.<\/p>\n\n\n\n

    Next, we want the APIs in our code to be reachable from the outside world via HTTPS. Install Caddy<\/a> the super simple HTTPS reverse proxy, edit the caddy conf file at \/etc\/caddy\/Caddyfile this way<\/a> and start it up with systemctl. Caddy now reverse proxies to localhost:8000.<\/p>\n\n\n\n

    couincouin.click<\/strong> {\n        root * \/usr\/share\/caddy\n        file_server\n        reverse_proxy localhost:8000<\/strong>\n}<\/code><\/pre>\n\n\n\n
    Launch your python virtual env – if you are not sure more info here<\/a> – and launch script.py. You should get something like that showing up in the terminal… <\/p>\n\n\n\n
    (back-end-env) ubuntu@instance-gpu:~\/backend$ fastapi dev backend.py<\/strong>\n[...]\n    server   Server started at http:\/\/127.0.0.1:8000\n    server   Documentation at http:\/\/127.0.0.1:8000\/docs\n[...]<\/code><\/pre>\n\n\n\n
    Last step is to attach my VM public IP to my domain as a A record – I will tell you why later – I end up with 4 APIs:<\/p>\n\n\n\n