Announcing, a humble little agent framework that I'm working on

Hello everyone! I’m Bret! I’m relatively new to the forum. I’ve done a lot of development work in my life, including creating open source modules for VCV Rack under the name Voxglitch.

I plan on releasing a new open source agent framework soon called It’s just like every other agent framework! Ha ha ha. Where it shines is the ability for agents to communicate with other agents.

I’ll post updates to this thread. I hope to make the code available in about a week. It needs a lot of work, but the foundation is looking pretty good.

Quick highlights:

  • Supports function calling (based on agent)
  • Supports RAG
  • Installable via Docker or locally, for faster development
  • Agents are all separate processes, allowing for potentially interesting features.
  • Originally I built this with the hope that agents could create tools and test them without having to restart the application. I don’t know if that’ll be possible, but it’s a fun goal!
  • Technology: Python, Node.js, React.js, RabbitMQ, Papertrail, MongoDB

Thank you to the community for all of the friendly help you’ve provide me so far. :man_bowing:



I just started job hunting, but I’ll try to keep working on the framework.
Here’s the code: GitHub - clone45/r1n: Multi Agent AI Framework

Thinking out loud…

Although I’ve gotten pretty far, I’m starting to reconsider the value of this framework in its current state. It works, but is it really useful? Maybe?

I’m thinking of focusing more on the individual agents, and less on making a holistic framework. A framework could later be built using these agents, but they would run fine individually.

Instead of using a message bus, an agent would interact with the outside world using HTTPS. An agent would provide a RESTful API. And the agent would be armed with tools, RAG, and other capabilities.

What would be the use case?

This would provide developers a good starting point for implementing an agent. Those agents could be used for website chat bots, as part of agent frameworks, or part of automations using Zapier or similar tools. They could be scheduled to run tasks, etc.

Why not just call OpenAI directly?

Agents would make function calling pretty simple. I’ve already created a drop-in tool system that I can reuse. This means, for example, if you want an agent to update a CMS, search for information using Perplexity, or process local files, this solution would be make it easier.

I’ve found that creating a framework that supports every LLM gets really complex. I think that it’s do-able, but I’m interested in treating agents more like microservices.

Anyhow, enough of my mental meandering! I’ll post more updates when I have them!