๐Ÿ„

Turn forgotten hardware
into a thinking network

mycoSwarm connects your machines โ€” old laptops, mini PCs, Raspberry Pis, GPU workstations โ€” into a single AI swarm. No cloud. No API keys. No data leaves your network.

# option 1: one-liner
$ curl -fsSL https://mycoswarm.org/install.sh | bash

# option 2: pip
$ pip install mycoswarm

$ mycoswarm chat
# that's it. you're running local AI.

What the swarm can do

๐Ÿ’ฌ

Chat with memory

Persistent facts and semantic search across all past conversations. Ask about something from weeks ago โ€” it finds the right session, not just the last ten.

๐Ÿ”

Parallel research

Plans multiple searches, distributes them across CPU workers, synthesizes a cited answer on the GPU. Faster than any single machine.

๐Ÿ“š

Document library

Drop files into a folder. The swarm indexes them and answers questions about your documents with citations. Local Perplexity.

๐Ÿค–

Agentic routing

The model decides when it needs web search or document lookup, shows you what it's doing, and uses the results automatically.

๐Ÿ”Œ

Plugin system

Drop a folder, restart the daemon. Your node advertises a new capability. No core code changes needed.

๐ŸŒ

Zero config discovery

Nodes find each other via mDNS. No IP addresses to configure, no config files to edit. Plug in and the swarm grows.

See the whole swarm at a glance

mycoSwarm live dashboard showing node status, memory usage, and task routing

Live swarm monitoring โ€” every node, every stat, real time.

A working 5-node swarm

Node Hardware Role Cost
Miu RTX 3090, 64GB RAM GPU inference โ€” runs 27B models ~$850
naru Lenovo M710Q, 8GB RAM Web search, file processing $50
uncho Lenovo M710Q, 8GB RAM Web search, coordination $50
boa Lenovo M710Q, 8GB RAM Web search, code execution $50
raspberrypi Raspberry Pi 2, 1GB RAM Search, lightweight tasks $35

Total: ~$1,035 ยท Zero monthly fees ยท Every node counts

A laptop on WiFi asked a question. An RTX 3090 on ethernet answered in 19 seconds. The weakest machine borrows the strongest brain.

Task flow

User asks question on Node A
  โ†’ Node A checks: can I handle this locally?
    โ†’ Yes: execute locally
    โ†’ No: orchestrator scores all peers
      โ†’ Dispatch to best peer
      โ†’ Stream response back to Node A

Research task on Node A
  โ†’ GPU node plans 4 searches
  โ†’ CPU worker 1 searches query 1  CPU worker 2 searches query 2
    CPU worker 3 searches query 3  CPU worker 1 searches query 4
  โ†’ GPU node synthesizes with citations
  โ†’ Stream answer to Node A
"If a student in Lagos with two old laptops can't participate, the framework has failed."

Named after mycelium โ€” the underground network connecting a forest.
It doesn't centralize. It finds what's available and connects it.

Keep reading

Every node counts

Got hardware collecting dust? Give it a purpose.