Ask HN: Open-source AI Agents building platform?

Hi,

There is a ton of tools comming out daily, mutating and pivoting in AI space. What I'm looking for is a platform to build internal agents that have way more permissions that you would dare to open to the customers. So maybe in the whole space , somebody has already done so and can point in the right direction.

Generally my idea is following: I'm building startup and my intention is to limit the number of support staff needed. Therefore I want to have specialised agents that does some functions for me. Primary communication is in slack - so intended workflow is that I write something to agents chat and they, having tools, do the tasks and get back to me in slack. It would be ideal to have those agents build on single system as oppose to coding each one of them separately. And since the intention is to give access to full data - self-hosted version is ideal, but open source would work just as well.

Anybody solving the same issues and can share relevant experience? Thanks a lot.

24 points

・

rkuodys

・

4 days ago

6 comments

lunarcave ・ 4 days ago

Doing something open-source [1] in the space. Self-hostable as well. Our customers actually have built internal things to help them with Zendesk / replace some manual Retool tasks.

There's a slack [2] integration as well. Fun fact, we use it to query our Zendesk / Postgres / Auth system from Slack and avoid internal tools.

E-mail in the profile. Happy to chat, even if it's to recommend something that works better for your use case.

[1] https://github.com/inferablehq/inferable

[2] https://docs.inferable.ai/pages/slack

  • rkuodys ・ 4 days ago

    Thanks I will check it out. Seems like it might be spot on what I'm looking for.

    But I have to ask - why do you hate Python? :))) I mean AI space has is python first (intuitively) so SDK for python would seem to be logical. Or am I missing smth?

    • lunarcave ・ 4 days ago

      We don't hate Python at all. We've just been waiting for someone to ask for it :)

      We're actually language agnostic (because we're operating at the level of the API). We're going after more underserved languages from the current AI tooling, and that was the reason we focused on languages other than python.

      With that being said, our ETA for a language SDK is ~24 hours.

dabe19 ・ 3 days ago

I use a combination of the crewAI python library, chromaDB as a vector database, and I run an embedding and llama model out of an Ollama docker container. This is all running and hosted locally on my personal PC and running off a 12GB RTX3060. I had to customize some of the tooling that comes with crewAI to make the RAG retrieval tool they provide work with my local chromaDB and Ollama models.

They have a nice class setup that lets you build fully custom tooling, really fun playing with the Selenium webdriver library and give your agents tooling to interact and observe web elements.

Huggingface recently released their smolagents framework as well, I might start looking into that.

overu589 ・ 4 days ago

You run a service that executes shell commands in whatever permission mode, and then you have your chat bot output their commands # and comments which the service then executes at your peril.

The handler that hands off between these pieces can post the stdout/stderr to chat.