Skip to content
MemMini

Quickstart

Same files, four depths. Start where you are; you can go deeper later without starting over.

Almost everyone starts the same way, by telling an agent to do it:

install MemMini from https://github.com/aaron-earle/MemMini

Your agent reads the install guide, asks you a few questions, and builds your mind from what you already have. It stays private: on your machine, or in a repo only you can see. Nothing about you is published unless you say yes.

The depths below are the install-depth knob: how far MemMini reaches into how you work. You don’t need any of it to get value (ask for an outcome first).

Chat

Section titled “Chat”

You live in AI chat, and you don’t have an agent that can install things.

  1. Open CONTEXT.md on GitHub and copy it.
  2. Paste it into your AI chat at the start of a conversation, or into custom instructions, a project, or a gem.
  3. Edit a line to match you (see Write your first rule).

Every reply now follows your rules and sounds like you: no syntax, ever.

Go deeper → tired of re-pasting? The next step makes it permanent, and an agent does the work.

Own

Section titled “Own”

You build with AI tools and you want this to stick.

Say the sentence at the top to any agent that can read a repo. What comes back is your mind: your own private repo, with the MemMini engine pinned inside it and your context/who.md and context/how.md written from your answers.

Your facts and your style now travel with every AI tool you build with, and the repo captures what changed after each session. Editing yourself is editing those two files.

Go deeper → want your agents coordinating through the repo, not through you?

Build

Section titled “Build”

You have a real workflow.

Point every agent at the mind you already have. They read and write the same repo, so they hand work to each other through git: coordination with no orchestration server to run. Be the boss, not the bus. (How work flows here walks one handoff start to finish.)

When a submodule helps: if your mind lives inside a bigger project of your own, mount your mind repo as a .mind folder in that project (git submodule add <your-private-mind-repo> .mind), same files, kept as a separate piece. The engine is never the submodule — it lives inside your mind, not beside it.

Go deeper → running more than one agent at once? Compile overlays per surface.

Lead

Section titled “Lead”

Fleets, frameworks, multi-vendor.

Terminal window
# your generated compiler (from wiki/contracts/compile-declaration.md — no
# compiler ships in the repo; scripts/ is empty on purpose)
<your-compiler> --overlay ./my-org --platform my-stack
# → CONTEXT.md (inject) + ENFORCEMENT.md (route) per surface

You get routed enforcement, cross-vendor agent coordination, and self-distilling memory: agents coordinating through one repo you own, you out of the middle.

The compile step layers content in order: framework defaults → your --overlay → an optional --platform layer. Same-name files replace; extra files append; your behavioral who.md is never pulled from a platform layer. Full model and flags: Repo layout.

Next

Section titled “Next”