Praxos: Kernel for AI Agents

Hey HN!

We're Lucas and Soheil, and we are the co-founders of Praxos.

Praxos is a kernel for AI agents, combining rock-solid stability with LLM flexibility.

The Problem: Building AI agents that reliably handle complex, stateful tasks is incredibly hard. Current approaches are often brittle, leading to unpredictable behavior and constant fires that have to be put down one by one.

Our Origin: We faced this when building AI for Insurance. LLMs struggled with the intricate, interconnected data, which as a bonus also wasn’t limited to a single domain. This made building a robust system very difficult, and connecting the different agents and workflows basically impossible. We knew there had to be a better way to combine AI's intelligence with system-level dependability.

How Praxos Works: We are building the full Praxos Kernel brick by brick, and we started with the most critical challenge: solving the data problem for AI, once and for all. Today, this foundational layer of Praxos is live and available through our SDK.

As of now, we can parse any data source, from unstructured PDFs and API streams to conversational messages to structured databases—and automatically transform them into a single Knowledge Graph with deep semantic types (e.g., 'PolicyID', 'CountryName') that allow you to ensure that you’re retrieving and updating exactly the type of info you want, while also supporting vector search. It also makes extracting structured information super easy, because all you have to do is look for the right types, no re-extraction necessary. Need new types on the fly? Not a problem. Simply create your own types, or extend ours through our pydantic compatible SDK.

Next Steps: We have laid out our vision and how we want to go about building this in our white paper, linked in the post. That said, the TLDR is that we are aware that a kernel is not just about managing memory. Our next goals will be to incorporate lifecycle management, syscall boundaries, and finally, what we are the most excited about: breaking down LLM tasks into simpler, repeatable instructions. Let us know if you think there’s a different step we should tackle first!

Try it Out & Share Feedback: Website: www.praxos.ai

Read our white paper at: https://www.praxos.ai/blog/ai-agent-kernel

Join our community at: https://discord.com/invite/2kmqTq7Yze We’re eager to hear your thoughts.