Local AGI - The local AI stack in one app anyone can use
by•
LocalAGI bundles the local AI stack into one desktop app: a team of AI agents, local models including image and video, an end to end encrypted private network, and an optional public network. Hive mind splits one model across machines you already own.
Replies
Hey Product Hunt, I'm Luke, the maker of LocalAGI.
The most advanced AI stack on earth (agent teams, local models, encrypted AI networks, distributed inference) has been free and open for years. Almost nobody uses it, because every piece assumes you live in a terminal, juggle config files, and know what an API key is. LocalAGI is that stack as one desktop app: download, type a task, review the result.
What's inside:
- A team of AI agents, not a single chatbot. An orchestrator routes each request to Browser (live web and automation), Image (generation and editing), Video (generation), or Hermes (a self-improving agent for everything else).
- Local models, including image and video, running on your own hardware. No API bills, no caps.
- A private network: an end to end encrypted mesh of your devices and people you trust. A weak laptop borrows compute from a strong machine, even across the internet.
- An optional public network where your idle machine can lease spare compute and earn.
- Hive mind, the newest piece: split one model across machines you already own. My test rig runs Qwen3.6-35B (4-bit, 128k context) on a 16 GB MacBook plus a 16 GB RTX 5060 Ti over one ethernet cable. Neither machine fits it alone. A running hive joins your network as a single node.
macOS and Windows today. Linux is next.
I'll be here all day answering everything: the stack, the privacy model, how hives work, the roadmap. Brutal feedback welcome, it's how this gets better.
A native mobile companion app would be huge here. Being able to monitor my agents on the go, kick off a task from my phone, or get push notifications when a long-running job finishes would make this feel like a complete ecosystem instead of something tied to my desk.
@dursunltxo Totally agree... a native mobile companion is next on our roadmap.
Today, you can already use Telegram on your phone to chat with your LocalAGI agent team, launch tasks, and stay connected while the desktop app handles the agents.
The native mobile app will take this further: you won’t need to keep the desktop app running. You’ll be able to manage agents, start tasks, and receive notifications directly from your phone, while connecting to other computers running local models through LocalAGI’s public or private network. A complete ecosystem is exactly where we’re heading!
been waiting for something like this honestly, ran it on my old desktop and macbook and the hive mind thing actually worked without me messing with configs. the privacy angle is what got me interested but the split model thing is what made me keep it installed.
@vedatmdxs That’s exactly why we built Hive Mind 🙌 Our showcase setup pairs a 16GB Apple Silicon Mac with a Windows PC running a 16GB RTX 5060 Ti to run Qwen3.6-35B (4-bit, 128k context). Neither machine can hold the model alone, but together they serve it as one private node. Really glad you got it working without fighting configs!
A small request from someone running this on a laptop: please add a true low-power mode that pauses the hive mind and drops the desktop app to under 2GB RAM when on battery. Right now it eats through my charge fast and the fan kicks in even when nothing is running. Even a simple "eco" toggle in the system tray would be huge for portable use.
@aylingullu82005 Thank you for the detail. First question back: which machine are you on, Mac or Windows? The two behave differently here and it changes what you are seeing.
On a Mac, macOS manages the model's memory. While the model sits idle the system can give that memory back to other apps, and the GPU and fan should only spin up while a task is actually processing. The one exception is if your machine is serving others in a hive or on the public network, which is real work even when your own screen is quiet. If neither is on and the fan still runs, that is a bug and I want to dig into it with you.
On Windows, today, a model stays fully resident from the moment it loads until the app closes. If that is your machine, the footprint you are describing is real, and it is exactly what eco mode has to fix.
Either way the eco toggle is a good request, and I would take it one step further than a tray switch: on battery it should kick in automatically. Eco mode would unload the model, pause the hive mind and network participation, and stop background agents, which is what makes a sub 2 GB footprint possible. The honest cost is that your first message afterward waits for the model to reload. On battery that is the right trade.
It goes on the list today. Until it ships, closing the app on battery frees everything, which is not the answer I want to keep giving. Reply with your hardware and I will tell you which case you are in.