Skip to content
MemMini carries one design choice across a phone chat, a desktop coding surface, and an agent fleet. A phone chat says make the main button blue; a desktop coding surface applies the blue primary button and notices the nearby button should match; a mixed fleet of surfaces carries that design choice forward. Primary buttons are blue.

MemMini

English is the programming language of AI. Every AI you touch learns a version of you. Your brain, their business. Your memory, their moat. MemMini takes it back. Fork you.
Plain-text filesUnder 1 MBNo service to runApache-2.0

MemMini keeps one digital you in plain English: synchronized, organized, and yours. Any AI can read it.

A bit of you here, a bit of you there, the whole of you nowhere. Every tool keeps a piece, and you’re the supplier and the subscriber.

The source code of you

Your voice, judgment, and preferences as plain-English files you own. One digital you, with the leverage to exist anywhere you need to.

Your own agent army

Deploy across a fleet of coding agents and harnesses. Carry the source code of you, memorized and synchronized. Be the boss, not the bus.

It gets smarter as you go

MemMini drafts its own next version after every session, sharpening however and wherever you use it, across every tool and subscription you have today or adopt tomorrow.

For seventy years, writing software meant writing code, a language only specialists spoke. Not anymore. English is the programming language of AI, and you already speak it. Say what you want in plain words and an AI builds it. That’s all MemMini is: software written in sentences. Your files aren’t notes about you; they’re the source code of you, and you’ve been a fluent AI programmer your whole life. The long version, and the research →

StartBuildExpand

Start

  • 📖  Read. Open the file and there you are, the source code of your digital self, in plain English. A self you can read, nothing hidden.
  • 🔒  Own. Right now your digital self is their asset, leverage they hold, value they bill you for. Fork it back and the balance flips: that self lives in your repo, locked to you, theirs no more. Fork you.Fork it

Build

  • ✏️  Shape. Change a line and you change who you are to every AI that reads you. Sentences, not syntax, no code to learn.
  • 🌱  Grow. MemMini drafts your next version after every session and waits for your yes. It gets sharper the more you use it, everywhere you use it. It writes itself; you own yourself.

Expand

  • 🧳  Move. Open any other AI: same files, one you, every tool, nothing re-explained.
  • ✅  Prove. Watch a rule actually work, and surface the moment it stops. Proof, not promises. Scale up and a whole swarm obeys one you.

One line in a file changes how the AI answers, and you can watch the rule work.

Source (context/who.md):

<!-- constraint: soft -->
- When a reply is drifting past three sentences for a yes/no question,
stop and lead with the answer.

Before

Q: Should I use a submodule?

That’s a great question! There are several factors to weigh here. Submodules pin an exact commit, which has tradeoffs around updates, and you’ll also want to consider subtrees, vendoring, and your team’s workflow before deciding…

After

Q: Should I use a submodule?

Probably not. Fork it. A submodule’s only for when MemMini lives inside a bigger project of your own.

(Then the detail, if you want it.)

The rule didn’t just get added. MemMini checks that it actually moved the answer, and surfaces it when it stops working instead of letting it rot in the file. That proof is the whole point. Know it’s working →

MemMini is a lot of things at once: a memory graph, the glue between your tools, a coordinated swarm. Minus the bill, the lock-in, and the meter.

A memory graph

Typed links the AI walks to the fact that matters: the right thing at the right moment, engineered for how attention actually works, with the budget in plain sight. No index to rebuild, no infinite-window hand-waving.

Your memory, on tap

Your whole self, ready for any AI, on the subscriptions you already pay for. No account, no meter, no monthly bill.

The glue between your AIs

One set of your state, preferences, and tasks that your chat apps, coding tools, and agents all read. It connects your tools instead of replacing them.

A coordinated swarm

Agents hand off through your repo; nothing ships while anyone still owes a piece. No orchestration server, no platform, no meter. Be the boss, not the bus.

Git, hosted anywhere

Versioned, forkable, yours. GitHub is a place, not a requirement: self-host it, keep it private, run it offline.

Same six moves. Four depths. You never start over, you only go deeper. (You never stop being you.)

Chat

I don’t code Live in AI chat. Paste one file; every reply now sounds like you. Start →

Build

I build with AI Fork the repo; your brain lives in git, yours for good. Start →

Code

I code Still your fork, now your agents hand work to each other through the repo. Start →

Orchestrate

I run a fleet Compile overlays per surface into a full agent-first system. Be the boss, not the bus. Start →

Quick path. Star, fork, run:

  1. ★ Star the repo. Tell us to keep going.

  2. Fork it. You now own your AI’s brain.

  3. Run it. Clone your fork, edit two files, recompile.

Terminal window
git clone https://github.com/<you>/MemMini && cd MemMini
# edit context/who.md + context/how.md, then:
./scripts/compile-context.sh

Any AI that reads the result now knows you. If you can write a note to a coworker, you can write software now.


Own your digital self. Fork you.

Go deeper: the design doc · contributing · security

It writes itself. You own yourself.