there is a list of things it will not do after you say yes
it stops and shows you the exact action before it sends a message,
makes a payment, signs a document or deletes anything
that part is an approval, and you can give it
below that there is a floor, and approval does not reach it
it will not trade securities and it will not move money between accounts
a classifier refuses to pass a card number or a government id at all
health and finances move only when you say so in words
and it never directly sees your passwords or your tokens
the interesting part is not the list
it is that somebody decided which of your yeses should not count
284B inference just hit 30 tok/s over open WAN
iroh v1.4.1 shipped this week. Mesh-LLM v0.78.1 patched chunked-request admission.
9 open-source GitHub repos that turn idle GPUs into one P2P inference mesh:
32,304 combined stars. All permissive. 6 pushed in the last 8 days, stars as of Oct 22.
JOIN THE MESH
01. iroh ★12.8k: https://t.co/kAzV00PcGI -> QUIC transport, NAT hole-punching
02. rust-libp2p ★5.6k: https://t.co/mktYLSGVvI -> peer discovery and relay routing
03. kalavai-client ★222: https://t.co/UUI7F25gl3 -> pool spare GPU capacity
SPLIT THE MODEL
04. shard ★465: https://t.co/m2j9Jd01t7 -> stream layer blocks across WAN
30.15 tok/s on 284B, six RTX 5090s, shard README
05. distributed-llama ★3.1k: https://t.co/TBMx3aDH8e -> split weights across home devices
06. prima.cpp ★113: https://t.co/8LiNqoWpen -> 70B on laptops and phones
674 ms vs 10,120 ms, 15x over llama.cpp, prima.cpp README
ROUTE REQUESTS
07. mesh-llm ★3.5k: https://t.co/vN6PutLXw6 -> one OpenAI API over pooled GPUs
08. pooled ★604: https://t.co/me9Bmk0vZU -> P2P inference in the browser
09. gpustack ★5.9k: https://t.co/rOoRQAWLZe -> schedule models across GPU nodes
the mesh:
peer → discover → relay → pool → shard → route → schedule → endpoint
curl -fsSL https://t.co/1Mk9VM7fGs | bash
Not a bigger GPU. A mesh of whatever you already own.
Most nodes alpha. Prompts travel open wire unless the mesh is private ⭣