cylinderthought

Is llama 3.3 enough? I think Gemini better

Nice work. I would take a look at x402 to accept payments directly with stablecoins and goes directly to sellers wallet. Easy to do. Most landing domain sucks.

do a B2C with a B2B branch

works best

facts

haha

me

tremendous

comments welcome :)

big fan of vaultwarden (i self-host it).

the main difference is architecture: shardium splits the secret (shamir) so the server never holds the full data, whereas vaultwarden holds the full encrypted vault.

also the "grandma factor": getting non-tech family to setup a vaultwarden account was friction. this is just "find paper in drawer, scan qr".

100%. i actually use it for my gpg keys too. i need to make that clearer on the landing page so it doesn't look like a "crypto-only" tool. recovering ssh/fido2 backups is a huge use case. thanks for the feedback

running 4-of-2 definitely removes the service dependency.

you highlighted exactly the "bug" i wanted to patch though: the collusion risk. if beneficiary A and B have a beer together, you get rugged.

the "live service" here acts purely as a time-lock. beneficiaries can't collude to rob you today because the 3rd shard isn't released until the dead man's switch actually triggers. it protects you from your own friends.

any feedback welcome! just a demo now

good.

the ability to satisfy needs with the least struggle

mcp. mcp. mcp

in google ai studio

my other cofounder is my cousin

make a discord

you can just do things

* dodo payments

* polar.sh

Communication is the most important pillar

langgraph’s killer for llm-savvy teams needing complex, stateful workflows, think multi-agent systems or cyclic graphs. it’s faster than custom pipelines, with langsmith debugging and one-click scaling. but it’s steep to learn, and debugging multi-agent setups can suck if state’s messy. langchain’s better for simple, linear tasks; langgraph’s for intricate control. autogen’s easier but less precise; crewai’s rigid. langgraph’s your pick for dynamic, production-ready projects.

for me it's less about dopamine crash and more about clashing headspaces. coding is pure flow, all in. ai pulls you out to play editor, and that shift kills momentum. here's what i've tried:

* batch ai work: toss it small stuff like snippets or lookups while i stay deep in my zone.

* hard limits: ai handles grunt work, not the big logic. if i'm fixing its bugs faster than writing it, i skip it.

* async vibes: treat ai like a junior coder. send a prompt, do something else, check later. no staring at the spinner.

if ai kills my dev job id switch to brewing craft coffee or restoring vintage synths. both need human finesse ai cant match. id lean on my debugging precision and patience. people will always want good brews and retro sounds.

this approach is cool for speed and small teams but duct-taping features with low/no code tools quickly gets messy. you lose the deep control engineers provide and debugging complex flows turns into nightmare mode. also testing? good luck with automated tests when your logic is scattered across 5 different services and zapier chains. scaling this needs discipline or you’ll end up with a brittle spaghetti monster nobody wants to touch.

it’s a neat hack but definitely not the new normal for anything beyond MVPs or tiny startups.

minecraft ui shows how much you can do when you actually have limits. pixel art forces clever shading tricks to make buttons pop and show depth without wasting pixels. meanwhile modern “glass” ui just slaps on blur and calls it a day. games need info fast and clear or you’re dead, apps could learn a lot by stealing from this “do more with less” mindset instead of making everything look like a soap opera.

big goals are always fuzzy and overwhelming. what works for me is breaking them down backwards, start with what success looks like, then ask what step comes right before that, keep moving step by step till you reach today. it turns the giant scary goal into small clear actions. also, keep revisiting and adjusting as you learn more. simple, effective, and helps avoid paralysis by analysis.

frameworks like GTD or PARA help but none beats clear incremental steps you trust.

honestly having unique stuff like torrent infra shows you’re hands-on and creative. it’s a convo starter and proves you can manage complex systems. just be ready to explain why it matters to your job goals. employers love people who can build and automate real things, even if they’re a little unconventional.

1) secrets in code are a symptom of broken dev culture not just tooling. fix the culture first, train devs early on using environment variables and iam roles. make secrets invisible to code.

2( relying on gitignore or vaults alone is band-aid. sdk picks up creds from environment or home dir automatically if you use aws properly.

3) automate secret scans in ci but dont trust them blindly, human reviews and rotating keys asap is still critical.

4) biggest risk is devs rushing and skipping processes, so build workflows that make mistakes obvious and costly to push.

5) at scale, even perfect tech fails without good process + education. focus there and tools become backup not main defense.

this mindset saves time and downtime. secrets leaks cost more than any fancy tool subscription.

i think what you want is a build step that acts like a “llm macro”, like you write comments or specs in the code, and the ai fills in the implementations automatically. its like literate programming but powered by llms.

theres a big gap now between interactive chatbots and fully automated code gen at build time. so a tool that integrates with the build pipeline and produces documented code would be a game changer.

also if it explains why it added code not in your specs that would improve trust a lot. this is possible with current models and some clever tooling but nobody cracked the ux yet.

personally ive tried a few hacks with rust macros and cargo steps to prototype this and it works ok but still rough around edges.

this will definitely be a hot area in next 1-2 years