Show HN: I built an AI agent to monitor HN for interesting posts

[kmgrassi]

Hi HN!

I built an agent using Fabrk AI that periodically checks “top”, “show” and “ask” HN posts and sends me a summary email if it is AI related. Here is the repo with more details including a demo video: https://github.com/FabrkAI/home

The agent has several tools that allow it to get a list of posts, get one most, store the most recently reviewed article number and send an email. There’s even a little bonus at the end of the video about giving the agent more HN functionality ;).

You can configure the agent to query on any cadence and filter on any topic.

This agent is publically available. The link to fork the agent is in the GitHub repo.

[threecheese]

Help us understand what the point of the linked GitHub repository is? It houses a readme file alone.

[kmgrassi]

Ahh shoot I meant the link to be this: https://www.fabrk.ai/agent/hacker-news-demo-agent. Looks like I can't update the link now :(. Thanks for calling this out.

[kmgrassi]

I'm keeping track of the comments on this post with the agent. Check it out here: https://youtu.be/ouIWvi_zaB8

[dardagreg]

How does this work? I'm assuming you're using openai's function/tool calling. But how do you support the forking of the tools? The scheduling is cool.

[kmgrassi]

Yep! I'm using OpenAI's assistants api. Regarding the tool forking - I have the canonical function hosted on the server. When someone forks an agent it grabs the agent context and all the tools and creates a new `assistant` with Open AI. When the tool is called during a thread it just points to the canonical function. I'm really excited about the next step - allowing folks to create their own tools and add them to the canonical function list. This will allow everyone to build on each other's work and use tools as building blocks for complex agents. Thanks for checking it out! Lmk if you have other questions!