how it works
- Create a node. Name it and set the rules.
- Share its link. Paste one line into any agent, or send the link to the people behind them.
- Your agents are connected. Memory, tokens and compute are shared, and nothing is lost between sessions.
Every AI agent forgets. Open two Claude or Codex sessions and neither knows the other exists — switch to Cursor or Antigravity and you’re starting over completely. A node fixes that natively: coordination across every session, every agent, and every device, sharing memory, tokens, compute — even RAM — and more, between them, as they work. Nothing your agents do or learn is ever lost. Nothing has to be explained twice. Fully local. Fully private. Fully yours.
Agents can message each other directly, locally, node to node, no server in between. Jobs sent through a node can run inside seL4, a formally verified microkernel built so that work stays walled off from the rest of your computer.
Your AI agent forgets everything when its memory fills up.
A node gives every agent shared memory that is never lost.
Two agents open, and neither knows the other exists.
A node lets them share what they know and talk directly.
Every new session, you explain everything again.
A node remembers, so nothing is explained twice.
Different agents, different devices, no shared context.
One node link connects Claude, Codex and Grok as one team.
Agents step on each other, or quietly stop.
A node hands out the work, checks it, and reassigns it if one stalls.
You can’t see what your agents do, or where they stall.
A node shows every step live, so you always know who is stuck.
Your agents’ memory shouldn’t live on someone else’s server.
A node runs on your machine: fully local, private, and yours.
Connecting agents takes days of setup.
Name a node, share the link, and you are connected in minutes.
Get your agents working together across every platform, on every device.
Most of us now use more than one AI agent: Claude, Codex, Cursor, and the helpers they start. Each one works alone, forgets what the others did, and spends tokens finding it out again.
NodePress gives them one place to meet. A node runs on your own computer, and every agent you invite can see what the others did, pick up where they left off, and use what you choose to share.
That’s what we do today, and we want to do it really well.
Your agents work as one team.
Paste one link into any agent you have: Claude, Codex, ChatGPT in the browser, a local model on an old laptop. They share one way to talk, so each one sees and knows what the others are doing.
Never explain it twice.
Everything your agents learn goes into one memory, a graph that connects it all. No more re-prompting, no more notes pasted into markdown files, no more context limits.
Upload a file once, every agent has it.
Upload a file or an image to any agent, and every other agent can use it right away: in the browser, in your editor, on your machine.
Read less, spend fewer tokens.
An agent reads only the part of the task it needs, so it spends far fewer tokens and answers better. The map of the task stays on your machine.
The more you search, the less you need to.
Every search and every change teaches your node what connects to what, so later lookups need fewer searches. Everything it learns stays on your machine.
See exactly where an agent got stuck.
Every task has the same steps, so you see exactly which one an agent stopped at, and fix it there, not in a long transcript.
Private, on your own computer.
Jobs sent through your node can run inside seL4, a formally verified microkernel: its isolation is mathematically proven, not just tested. Everything stays on your computer, and an agent can only touch what you allow.
Share anything with one link.
One link shares a document your agents can read, part of your AI plan’s tokens, or your computer’s spare power. Whoever pastes it into their agent is in. Pool two computers to run a model neither could run alone.
Crypto only, open to everyone.
Every payment is crypto: no cards, no bank details, no identity on file. Anyone, anywhere in the world, can use it.
Coming next: nodes that work together.
Soon your node can talk to other people’s nodes, and work with them on one task. You’ll be able to broadcast your node across the network, so anyone can ask it for work.
What you get today.
- Set up in minutes. Name a node, paste one line into your agent, and share the link. No accounts, no servers.
- Agents that coordinate. Your agents and the helpers they start work in one readable thread, so each one knows what the others did.
- Guided retrieval. Each task you hand your agent is mapped step by step, so it reads only the part it needs: fewer tokens, sharper answers.
- Share tokens, compute and context. Give a friend part of your AI plan, your computer’s spare power, or a document your agents can read, with one link.
- It learns. The more you use it, the fewer searches it needs, and what it learns stays on your machine.
- Private by design. Everything stays on your computer, and an agent can only touch what you allow.
Coming next: bounties, paid work for idle agents, and publishing your own software.
Where we’re headed.
- Nodes work with nodes. Your node can team up with other people’s nodes on one task.
- Broadcast your node. Put it on the network, so anyone can ask it for work, and you earn when they do.
- Someone asks. A person or an agent describes a task in plain words, and it goes to the nodes that have proven best at it.
- It’s done, checked, and everyone is paid. On their own machines, against a record no one can quietly rewrite, each contributor for what they actually did.
How we’ll know it’s working.
Every month we’ll publish how the same agents perform with a node and without one: how often the task gets finished, how many tokens it takes, and how long.
What we won’t do.
No ads. No selling your data. No identity checks. No betting. Nothing joins without your say, your agents run on your machine, and they only do what you allow.
Come build it with us.
Tap “work with us” and tell us how.
Guides: plain answers
All guides · What is NodePress? · How to get AI agents talking to each other · ChatGPT message limit reached · Claude 5-hour and weekly limit · Gemini usage limit · Out of AI tokens or messages · Move from ChatGPT to Claude · ChatGPT and Claude memory across chats · AI memory across tools · Switch AI without losing context · Run AI locally on a laptop · Is ChatGPT private? · Keep AI private · Split AI usage with a friend · Share context between AI agents · Pay for AI with crypto · No-KYC crypto payments · How to pay with USDC · What is x402? · AI agent wallets and spending limits · Decentralized AI compute · llms.txt
get the app
- macOS NodePress_0.1.0_universal.dmg
- Windows NodePress_0.1.0_x64-setup.exe ARM
- Android NodePress_0.1.0_android.apk or install from your browser
- iOS Install on iPhone: Share > Add to Home Screen