Every session I had to re-explain everything. There are .md files and memory systems but you have to manually tell the AI to use them, and when you're actually deep in work you don't think about that. They're supposed to be smart. Why can't they just learn?
I tried building self-learning systems on top of .md files for a while. Never really worked. Then I thought: I can remember all of this naturally. Why can't they? There has to be a way to build this.
So I started researching how human memory actually works. The brain is genuinely stunning. I ended up modeling 124 of its mechanisms in TypeScript. Spreading activation, Hebbian learning, connections that strengthen every time they fire together.
Still early (v0.1.5) but already working really well for me. Local MCP server, SQLite, no cloud. npm install -g veris-dev
Yeah, the title is a bit dramatic. Basically it's spreading activation + Hebbian learning modeled as a local MCP server. Works across editors, fully local.
Every session I had to re-explain everything. There are .md files and memory systems but you have to manually tell the AI to use them, and when you're actually deep in work you don't think about that. They're supposed to be smart. Why can't they just learn? I tried building self-learning systems on top of .md files for a while. Never really worked. Then I thought: I can remember all of this naturally. Why can't they? There has to be a way to build this. So I started researching how human memory actually works. The brain is genuinely stunning. I ended up modeling 124 of its mechanisms in TypeScript. Spreading activation, Hebbian learning, connections that strengthen every time they fire together. Still early (v0.1.5) but already working really well for me. Local MCP server, SQLite, no cloud. npm install -g veris-dev
Uh huh. Sure.
Yeah, the title is a bit dramatic. Basically it's spreading activation + Hebbian learning modeled as a local MCP server. Works across editors, fully local.