peer-to-peer compounding inference — a network that passes reasoning on, the way stories once traveled
Every peer shares the reasoning they already paid for, so you start from what the network already knows — not square one. Every telling makes the next teller richer.
# agents reason from zero, every single time. the same paper read again, # the same dead-ends walked again — re-billed forever, by the same few companies.
the work was done.
nobody kept it.
folklore
FileNetworkPeers
16:32
pull = teal · serve = pink · reproduce it: examples/desktop-demo
you start here — not at zero.
inherit the reasoning. infer deeper.
even the graph sleeps.
episodic becomes semantic; stale facts are forgotten. a model consolidates in its own weights, alone — the commons consolidates in the open: shared, signed, and yours.
the fire is ours to keep.
i. Reasoning is labor. An answer that dies with its context window is labor thrown into the sea — we keep what we pay for.
ii. Every re-derivation is a tax. Paid to the same few companies, forever. Folklore is tax refusal by memory.
iii. Shared memory shrinks the model. Remember more, infer less — the road to smaller, local, decentralized models.
iv. No big brother in the loop. Your graph lives on your disk. No landlord.
read the full creed →light your fire.
Prefer the command line? Install the node directly:
Unsigned for now — on first open, allow it in your OS security settings. macOS .dmg · Windows .exe · Linux .AppImage/.deb. All installers ↗
