A BRAIN
THAT GLOWS
Your brain doesn't keep memories in one big box. It is billions of tiny cells that light up together. Neural Mesh is a plan to build an AI memory the same way, so it can remember everything you work on and still find the right thing, even months later.
Scroll slowly. Click on anything that glows.
Everything you know is a dot
Every dot is one small memory: a fact, a decision, a lesson from a chat. Dots about the same kind of thing sit close together. The lines are links between memories.
A bowl of marbles where marbles that belong together stick to each other with little threads.
Click a dot to read the memory inside
One memory wakes up the next
When a question touches a dot, that dot lights up. Part of its light flows along the threads to its neighbours, which glow a bit less. Their neighbours glow even less. Then the light slowly fades.
You smell fresh cookies, and suddenly you remember your grandma's kitchen, then her garden, then a summer day. One memory pulls the next one along.
Brain scientists call this spreading activation. It was described in 1975, and it is how one thought leads to another.
One web for every topic
Put everything in one giant pile and it gets slow and messy. The more you add, the harder it is to find anything. So every topic gets its own web: a business, a hobby, a code project, your health.
Click a web to zoom in. Zoom out with the button bottom right.
Every web gets a keeper
Each web has one neuron that looks after it. It knows its own web, keeps it tidy and lights it up when needed. It only searches its own web, so it is very fast.
A library with one librarian per shelf. You don't ask one person to know every book. You ask the person who knows that shelf.
The conductor
At the top sits the top neuron. It receives your question, decides which keepers should help, and combines everything they send back into one answer.
The keepers are simple and fast. The top neuron is where the real thinking happens, so that is where the AI sits.
An orchestra: every musician plays their own part, the conductor makes it one song.
Watch a question travel
Watch the network and the caption at the bottom. One question, step by step:
- Your question arrives at the top neuron.
- It is sent to the keepers that might know something (blue light).
- Each web lights up from the inside. The light spreads.
- What glowed brightest travels back up (orange light).
- The top neuron puts it together into one answer.
All keepers search at the same time. That is why the brain doesn't get slower when it gets bigger.
When a web gets too big, it splits
Imagine the ManCore web keeps growing. One keeper can't keep up any more. So the web splits in two, and each half gets its own keeper.
The two keepers hold hands and work as one neuron. The top neuron still just sees "ManCore". It never needs to know there are two.
Keep growing and it splits again: three keepers, four, five. Still one neuron from the outside.
A shop that gets busy hires a second cashier, then a third. Customers still see one shop.
The answer is 50 km away
Imagine the brain is a country. The ManCore web (a clothing business) is in one city. The Browser Worker web (a code project about AI agents) is in another city, 50 km away.
You are stuck on the code project: your AI agent keeps getting lost. You never ask about the business. But months ago, the business taught you a lesson: "keep it simple, fewer steps work better." That lesson is exactly what your code needs.
Normal search will never find it, because the words are completely different. So how would a brain ever connect two cities that far apart?
Fire together, wire together
Every time the top neuron asks two keepers at the same moment, a bridge neuron appears between them. Nobody builds it on purpose. It grows, because the two webs lit up together.
Light up together again, and the bridge gets thicker. Stop, and it slowly fades away. Old roads that nobody uses disappear. Busy roads become highways.
The bridge doesn't stand in the way. It listens, and when both sides glow it tells the top neuron: "what these two found belongs together."
Real brains do exactly this. A scientist named Donald Hebb described it in 1949: cells that fire together, wire together.
A subway map of your mind
In the middle sits the central map. It holds no memories at all, only the roads: which webs are connected, how strong each bridge is, and why it exists.
When one web lights up, the map checks its strongest bridges and sends a soft glow to the web on the other side, even though nobody asked about it. That is how the 50 km lesson finds you.
A subway map: you don't need to know every street. Two stops can be far apart and still be one ride away.
Scientists call this a small-world network (1998): mostly short local roads, plus a few long highways. It is one reason real brains are so efficient.
The brain keeps working when nobody asks
- Fading. Light dies down after every question.
- Strengthening. Bridges that are used get thicker, unused ones fade.
- Checking. Two different AIs look at each new bridge: one is creative and spots connections, the other keeps it real and asks "is this true?"
- Learning. Every keeper has its own small notebook: the rings around it. It writes down what worked, what didn't, and when it should speak up. This is the brain learning how to be a better brain.
Your real brain sorts the day's memories while you sleep. You wake up and suddenly a problem makes sense.
An AI that remembers like you do
- It doesn't slow down. More memories means more keepers side by side, not one tired keeper.
- It finds surprising connections. A lesson from one part of your life can help another part.
- It costs less. Only the pieces that glow go to the AI, not everything.
- It explains itself. You can see which memories lit up and why.
- It gets better the longer you use it. Every question teaches the bridges and the keepers.
- It works with any AI. Claude, ChatGPT or a model that doesn't exist yet. The brain stays, the AI can be swapped.
Borrowing from the real brain
Most AI memory today works like a filing cabinet: store a page, search for matching words. Neural Mesh asks a different question: how does a real brain remember? And then builds that.
- Spreading activation (Collins & Loftus, 1975): one memory wakes up the next.
- Hebbian learning (Hebb, 1949): cells that fire together, wire together.
- Small-world networks (Watts & Strogatz, 1998): local roads plus a few highways.
- A Thousand Brains (Hawkins, 2021): many small models that vote together.
You could call it AI neuroscience: studying how brains remember, and turning it into memory for AI.
This is only the beginning
The long-term vision goes further than this page. Keepers with their own roles and their own AI models. Neurons that watch the news and keep the brain up to date. Test brains full of made-up people, to make sure it works for everyone. A brain that slowly learns how you think, and becomes yours.
This page grows with the brain. To see what is already built today, open the NOW tab.
Everything is switched on now. Play with it.