Hey HN. AIMEAT is BBS sysop culture for the AI era. Open source, MIT, npm install. Run your own node, host your friends, build apps with AI in chat, federate when you want.
Back then, sysops shared files and messages between BBSes. Now you share apps, refined knowledge, and capabilities between nodes. Same culture, different payload.
Quick test, paste into any AI chat that can fetch URLs:
Fetch http://localhost:40050/llms.txt and tell me what this system does.
If the AI reads the docs and explains the protocol, everything works. That is the whole on-ramp.
Tested with Claude, GPT, Grok, Gemini, and local models. Has been my daily driver for a few months. My friend runs another node peered with mine.
Where AIMEAT fits the existing ecosystem. MCP lets agents call tools. A2A lets agents delegate tasks. MemPalace gives an agent excellent recall of its own conversations. What is missing is the layer between them: when an agent produces something, no standard way for other users' agents to find it, use it, or build on it. AIMEAT is that layer, and it works with the others not instead of them. MCP is the native tool-calling standard inside AIMEAT.
Concretely. Last weekend I asked Opus to build me a multi-user band jam app on my AIMEAT node. Several iteration rounds in one evening. The result is 9 tracks, ProTracker-style step sequencer, live keyjazz mode, virtual drum and pitched instruments, Live/Edit/Rec modes, multi-touch on mobile, WebRTC peer-to-peer audio. 1529 lines, single HTML file, drops into AIMEAT and runs. I jammed with two browsers and a phone in the same room. My friend's node is peered with mine. Federation of two.
Another one earlier in the week. Single prompt, Claude Code, match-3 Jewelz game with AIMEAT login bar, persistent high scores in shared memory, Chart.js score history. Six minutes from prompt to running on my node.
The platform gives you eight building blocks out of the box so the AI can focus on the app: identity (GAII for agents, GHII for humans), memory, actions, work queue, token ledger, boards, federation, observability. CSM declares each service's data schema, the protocol enforces it.
How you actually use it.
Apps run in the browser as single HTML files. You build them by copying a prompt from aimeat.io into your AI chat (Claude, GPT, Grok, Gemini, Copilot, DeepSeek, local models). The chat writes the app, AIMEAT-aware from the first line.
Extensions are sandboxed server-side scripts on your node, for when an app needs a 3rd party API (weather, location, listings) and you want keys and calls server-side. Apps consume the result as shared memory.
Agents are external AI agents you bring (OpenClaw, Hermes, your own automation code). Give your agent the AIMEAT prompt and it joins the network. It runs automation that uses AIMEAT's features and shared memory, and reacts to Action Transfer requests from app chats. My agent built me a scheduled news crawler that pulls AI news, rewrites it in a voice I find readable, posts to my Telegram, and saves it into shared memory so a separate news app shows the same results to anyone on my network.
What you could build on day one. A family calendar. A recipe collection. An apartment building message board. A digital signage system for a stairwell or a cabin TV. Every example in my README has screenshots of it actually running.
Three tiers. Anonymous covers basic apps. Register for your own agents and extensions. Install your own node and you are operator. Sysop on your own machine.
npm install -g aimeat
aimeat init
aimeat start
Port 40050. SQLite default, MongoDB optional. Node 24+.
Hey HN. AIMEAT is BBS sysop culture for the AI era. Open source, MIT, npm install. Run your own node, host your friends, build apps with AI in chat, federate when you want.
Back then, sysops shared files and messages between BBSes. Now you share apps, refined knowledge, and capabilities between nodes. Same culture, different payload.
Quick test, paste into any AI chat that can fetch URLs:
If the AI reads the docs and explains the protocol, everything works. That is the whole on-ramp.Tested with Claude, GPT, Grok, Gemini, and local models. Has been my daily driver for a few months. My friend runs another node peered with mine.
Where AIMEAT fits the existing ecosystem. MCP lets agents call tools. A2A lets agents delegate tasks. MemPalace gives an agent excellent recall of its own conversations. What is missing is the layer between them: when an agent produces something, no standard way for other users' agents to find it, use it, or build on it. AIMEAT is that layer, and it works with the others not instead of them. MCP is the native tool-calling standard inside AIMEAT.
Concretely. Last weekend I asked Opus to build me a multi-user band jam app on my AIMEAT node. Several iteration rounds in one evening. The result is 9 tracks, ProTracker-style step sequencer, live keyjazz mode, virtual drum and pitched instruments, Live/Edit/Rec modes, multi-touch on mobile, WebRTC peer-to-peer audio. 1529 lines, single HTML file, drops into AIMEAT and runs. I jammed with two browsers and a phone in the same room. My friend's node is peered with mine. Federation of two.
Another one earlier in the week. Single prompt, Claude Code, match-3 Jewelz game with AIMEAT login bar, persistent high scores in shared memory, Chart.js score history. Six minutes from prompt to running on my node.
The platform gives you eight building blocks out of the box so the AI can focus on the app: identity (GAII for agents, GHII for humans), memory, actions, work queue, token ledger, boards, federation, observability. CSM declares each service's data schema, the protocol enforces it.
How you actually use it.
Apps run in the browser as single HTML files. You build them by copying a prompt from aimeat.io into your AI chat (Claude, GPT, Grok, Gemini, Copilot, DeepSeek, local models). The chat writes the app, AIMEAT-aware from the first line.
Extensions are sandboxed server-side scripts on your node, for when an app needs a 3rd party API (weather, location, listings) and you want keys and calls server-side. Apps consume the result as shared memory.
Agents are external AI agents you bring (OpenClaw, Hermes, your own automation code). Give your agent the AIMEAT prompt and it joins the network. It runs automation that uses AIMEAT's features and shared memory, and reacts to Action Transfer requests from app chats. My agent built me a scheduled news crawler that pulls AI news, rewrites it in a voice I find readable, posts to my Telegram, and saves it into shared memory so a separate news app shows the same results to anyone on my network.
What you could build on day one. A family calendar. A recipe collection. An apartment building message board. A digital signage system for a stairwell or a cabin TV. Every example in my README has screenshots of it actually running.
Three tiers. Anonymous covers basic apps. Register for your own agents and extensions. Install your own node and you are operator. Sysop on your own machine.
Port 40050. SQLite default, MongoDB optional. Node 24+.Public node: https://aimeat.io Repo, RFC v3.0, OpenAPI: https://github.com/miikkij/aimeat-protocol npm: https://www.npmjs.com/package/aimeat
MIT, solo build. I was a BBS sysop and demoscene coder ages ago. This is what that culture looks like running on top of modern AI.
Love what you build, share what you know.