# relays are now more reliable
or should i say relayable?
nostr.mom switched from my manual years-long algos to vibed replacements. for years we didn't have a fast write-policy script. now we have go code that does quick processing, flood protection etc (system 1 thinking) with help of python (slow but careful processing, system 2 thinking). python works async and feeds its findings to go.
most behavior is similar to previous but we omitted PoW ladder (harder and harder PoW requirements while client keeps flooding) for simplicity. those may return later.
nsfw and spam checks were off for a while, my GPU at home was not available. but those checks returned, and you may see less nsfw or spam on nostr.mom. fast go may slip a few bad events but slow python will correct it and future events from same IP should be throttled. the bad event that was accepted wont be deleted though. maybe i can do this in the future.
nos.lol write policy untouched but for sure this is coming to nos.lol as well.
turns out LLMs are pretty good at nostr relay operations. the REQ spam attack is still ongoing, 2000+ bots are now banned. strfry PR is on github. if you are a relay op and not using LLMs, consider rewriting scripts to vibed versions. i am now considering full agentic operation of the whole machine.
not open sourcing because it may allow attackers to learn weaknesses, but i can share private github with friends who are interested.
nevent1qqsy9g823ep78vjc08udykn93vs8a3l3vfp9snuqac00d6vxjexkzggl7fy7w
