smashah

thank you for saying this. We have to bring back shaming in tech.

Hi all,

I came up with cve.earth (https://github.com/cve-earth/.github/blob/main/profile/READM...) a few years ago as a standard way (based on CVEs in tech) to track environmental destruction for research purposes (not a place for whistleblowing). It is basically a Github Issue template:

https://github.com/cve-earth/disclosures/issues/new?template...

If you are interested in standardising this space and doing something to save our planet, please get in touch

https://txta.dev/smashah

Why can't we set the harness? Seems like the missing piece

How much did the Gulf states pledge towards this?

The UN needs a seal team 6 to take out everyone at the pentagon as it's home to the worlds biggest and most babybloodthirsty terrorists

You do not do all this with google flights. The link can be constructed with protobuf.

I was expecting Occam wearing a suit.

a marketplace for people to safely collect their own AI sessions in their own S3 compat buckets so they can sell them later to dataset buyers.

https://traice.market

499 connections :(

You can get covers that shield/protect the hinge

Pretty sure that's due to the tube.

D.C is controlled by enemies to humanity and freedom. They openly plan on enslaving the rest of the world to secure their hegemony.

Coming up soon is

npx txtadev inbox

and

npx txtadev read

Hi all,

I made peardrop.fyi because I have now perfected the agent-driven software factory stack. Now I only interface with an agent called Pennyworth that uses an aviation/Gotham/billions based cast/operative org structure to get stuff done. This means that sometimes Pennyworth determines that 5 projects are blocked due to lacking credentials/secrets/keys/JSONs etc.

Sometimes I talk to it via discord, whatsapp etc and therefore I would prefer that none of these secrets make it anywhere in the chat. I also don't want it to be going through the model routers, OTel, model providers, etc etc.

so this is why I made peardrop.

Peardrop uses pears SDK/HyperDHT a cli and an optional (default) relay (self hostable) to enable a pleasant and secure experience for transferring secrets/files to your agent's target location no matter where it is, no matter where you are.

The agent is instructed to create a declarative UI for the credentials request using toml and markdown. This allows you to have all the context/links/instructions you need in the peardrop link itself to actually find the right credentials and deliver it into the peardrop.

This ease of use for the submitter also makes it a good tool for requesting creds/secrets from your nontechnical clients.

The service is self-hostable, open-source and mainly free but due to the limitations of the web environment (no udp support), there is a hosted relay, and because I want to use every project as a learning experience, I have implemented x402 payments for using the relay for anything over 5mb - practically this means that nobody that uses it for it's actual use case (sending creds/secrets) will ever actually need to pay for the relay.

Here is the gh: https://github.com/smashah/peardrop

p.s if anybody has any work they need done, want to learn more about my agentic coding system, or want to get their vibecoded project over the line then please DM me. thank you.

(edit, oh wait there is no way to DM people on here, just make an issue in the gh or something while i figure out a way to send me a dm)

Mighty obsessed with slop, is that what you give Alex Karp?

Is it? Looks perfectly legible. Do people expect their forest-ignoring tree-nitpicking validated nowadays?

Please also make sure to keep your wikidata and ISNI information up to date with your substack links.

Services like pirate[dot]kred use that information to link your books to your payment pages (substack is supported).

Hi all, with the recent Grok discourse with them hijacking basically all your code, I wanted to create something to give everyone access to an easy way to collect their own AI coding session traces so that you (if you want, you don't have to) can sell them later in a way where it remains secure and within your control at all times.

Every AI lab subsidizes access to their models partly to get their hands on your sessions. In this AI goldrush, models are the shovels and our traces are the wood the shovel-makers build them from. So if our working sessions are that valuable, why are we giving them away for a mere discount to only one model provider at a time? I wanted to answer the question: what if we could sell our own traces to model providers directly in a way where the data is never at risk? That's why I'm building traice.market (WIP) and Traicer. Traicer is the open source collection service that runs on your machine, and you route selected Claude Code / Codex / OpenCode calls through and it saves to your own S3 compatible bucket (thanks alchemy.run and varlock for making integrating this feature securely very simple!)

Each supported call gets credential-stripped, redacted (keys, JWTs, DB URLs, emails), encrypted with AES-256-GCM, and written to an S3 bucket you own. Keys never leave your machine. When traice.market is ready, it will let you sell traces without us ever being able to touch or read them: the marketplace only sees signed content-free metadata (provider, model, token counts, hashes), and a sale requires your explicit approval, with delivery re-encrypted on your machine to the buyer's key.

I would love any feedback/questions etc.

thanks!

wonderful project thanks for making it :)

Use axiom all the time thanks for your work. The next frontier in iOS app tooling is self-hostable Linux based build server. Thoughts?

I'm working on pirate.kred - a Tip Jar DNS for pirates to directly and easily pay creators/authors.

I talked more about it here: https://news.ycombinator.com/item?id=48881942