Skip to content

To Chat #2

Description

@KenoticLabs

Hey Dominik, came across your Engram project. The spreading activation on weighted knowledge graphs with Rescorla-Wagner learning is a sharp approach, and the "privacy by architecture" principle (text never stored) is exactly right.

I'm building in an adjacent space a memory primitive called DTCM (Decomposed Trace Convergence Memory) that handles the "how do I remember" problem the same way you're handling the "what do I do with what I know" problem. Deterministic, no LLM on the read path, on-device. Two papers on arXiv (2604.06710, 2604.10981), 5 patents filed.

What caught my eye is how naturally these two layers stack. DTCM handles persistence and reconstruction across AI platforms. Engram handles reasoning and activation over the resulting graph. Both deterministic, both structural, both reject the "just throw an LLM at it" approach.

I'm building Kenotic Labs around making this the standard infrastructure layer. Small team forming, serious IP. Your Rust + systems background and the way you think about bounded domains vs LLM uncertainty that's rare.

Would be good to talk.

sam@kenoticlabs.com

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions