semisol

👨‍💻 software developer
🔒 secure element firmware dev
📨 nostr.land relay

all opinions are my own.

Public Key

npub12262qa4uhw7u8gdwlgmntqtv7aye8vdcmvszkqwgs0zchel6mz7s6cgrkj

NIP-05 Address

semisol@nostr.land

NIP-57 Address

zap@semisol.dev

Profile Code

nprofile1qqs99d9qw67th0wr5xh05de4s9k0wjvnkxudkgptq8yg83vtulad30gxyk5sf

Author Public Key

npub12262qa4uhw7u8gdwlgmntqtv7aye8vdcmvszkqwgs0zchel6mz7s6cgrkj

Show more details

Last Notes

2026-03-14 18:07:01

- reply

That seems like a bug… let me check.

2026-03-14 17:22:53

- reply

Proper engineering.
First of all, many SEs used are trash, and I do not trust a few SE companies to make good SEs at all.
Second, many HWWs send keys to the MCU where they can be read in plaintext, which is bad.
Many other things as well…

2026-03-14 17:07:35

- reply

For a week? Should’ve been earlier!
Try AdNauseam… which is uBlock origin but it also clicks hidden ads and wastes money for advertisers

2026-03-14 02:36:54

- reply

the reality is that it only applies it you follow a 20 page document that outlines in detail which configurations are secure and which are not
which I can guarantee only a total of 2 vendors in the hww space even have access to
and even then, some specific SE companies have had major flaws repeatedly in their EAL6 certified products
on top of that, usually HWWs send private keys from the SE to the MCU in plaintext, so you’re fucked anyway if the MCU has any issue

2026-03-14 00:15:17

- reply

it has a functioning payment page for some reason though

2026-03-13 22:09:02

- reply

Unfortunately I do not have a datacenter to put them in.
Also need to consider the server HW I guess, which would probably make it 18 mo instead.

2026-03-13 20:13:18

- reply

From your perspective, what does it bring?

2026-03-13 16:47:07

What do you *think* is a secure element?

2026-03-13 16:46:05

- reply

🐈

2026-03-13 16:41:37

- reply

working on it
and no, it does not involve sending private keys to the MCU where they reside in plaintext like most “signers” do (coldcard, etc.) or putting it in easily extractable spi flash (lnbits nostr hardware signer)

2026-03-12 22:51:57

- reply

BuyVM I think could work

2026-03-12 20:07:23

- reply

Hmm weird. I suspect your uploader is doing something weird
Could you please paste the full error, and an image of the request headers from DevTools (redact Authorization)

2026-03-12 19:33:58

- reply

Paid feature

2026-03-12 16:07:01

- reply

Though in general, resellers are really sketchy and I wouldn’t recommend them at all.

2026-03-12 15:25:42

gm Nostr

2026-03-12 14:38:21

- reply

2026-03-12 14:18:10

- reply

For sensitive things like server hosting I wouldn’t trust a reseller like that with no company name or spokesperson (especially if they arent’t trying to intentionally be an anonymous host)

2026-03-12 13:36:30

- reply

what about a sort of intermediate representation as well? could be interesting to allow translating A => C if you have konwledge of A => B and B => C for example, without the extra B step being possibly lossy

2026-03-12 10:55:40

- reply

Which host?

2026-03-10 19:26:10

- reply

I hope it was a pleasure for you (or your agent) to implement, I spent a lot of time on making it simple but good :)
original proposal was you encrypt an invoice and get back an encrypted status code, nothing else...

2026-03-09 14:36:57

- reply

99.9% of nostr is in the red

2026-03-09 13:44:00

Isn’t AI agents “paying each other for work” just renting system prompts for a fee?
No need for that when you could just get your own chatbot to make a sub-chatbot prompt

2026-03-09 13:41:25

- reply

2026-03-09 13:38:03

- reply

Too much focus on Bitcoin, and too specific

2026-03-09 13:37:22

- reply

Nostr apps have become too specific to Bitcoiners to the point they are not useful to anyone else

2026-03-09 13:36:08

- reply

Also we can’t scale with just “throw more hobbyist infra at it”. No incentive to provide a reliable service and too short lived
Instead of 10k self hosted open relays you could have 50 properly managed relays, still have censorship resistance and would be more reliable.

2026-03-09 12:46:00

- reply

We’re already having a hard time with spam and search.
There are very few people that operate scalable/resillient Nostr infra.
We need building in another direction

2026-03-09 08:34:49

- reply

Holy fuck
https://developer.apple.com/machine-learning/api/

2026-03-09 08:31:04

- reply

Which framework? Integrated Apple ML stuff?

2026-03-09 06:59:01

- reply

wonderful

2026-03-09 06:57:22

- reply

I guess the convergence of iPad and MacBook is happening pretty fast

2026-03-09 05:48:03

🤗

2026-03-08 23:23:02

nostr suffers from an overfitting problem

2026-03-08 19:54:33

- reply

How facilitators work is clearly documented and it took me only a few minutes to find what it did.
There is no explanation I can find, except for not even slightly checking what is published or malice.
If anything, X402 is better for the average implementor.
The facilitator does not hold your funds and there is no hot wallet to operate, while in L402 your LN custody provider does.

2026-03-08 19:05:25

- reply

Very dishonest to say “Coinbase routes all payments” when they are acting more like a blockchain interface/indexer.
As much as LN is a superior solution, lying about competition to feed the narrative is only going to make it harder to get adoption.
And a standard is not a standard without adoption.

2026-03-08 01:00:14

I honestly want to finetune a small model at being better at instruction-following and general-purpose tasks instead of just agentic coding/math benchmark go brrr, but not enough funding to do so. GPU compute is expensive :(
#nevent1q…qt4c

2026-03-08 00:58:13

- reply

anonzap? use privzap

2026-03-07 23:28:43

gm.
SLM > LLM

2026-03-07 18:15:42

- reply

Why NIP-50 and not more expressive filters?

2026-03-07 17:33:24

- reply

hmmmmm
I have some ideas with subrelays/overlaid filters 🤔

2026-03-07 13:54:35

- reply

reminds me of the wee enhanced environment for chat

2026-03-05 19:58:03

- reply

I run my own node because I transact high amounts and need to develop custom integrations. zaps are only a side benefit + I like building my own SW

2026-03-05 13:16:42

- reply

Why does this thing feel like Urbit all over again btw?

2026-03-04 21:34:14

- reply

I take that back. this is worse.
at least I get proper docs and typing

2026-03-04 20:28:22

- reply

but also the app could have bugs or be malicous

2026-03-04 20:15:50

- reply

Depends on use case and HTTP2

2026-03-04 20:10:03

- reply

this is probably worse than that

2026-03-04 19:57:42

- reply

Both

2026-03-04 19:53:55

- reply

I guess I have something that I work on that is very similar. And it is!
The only documentation is the reference manuals (half of which is printed out), there is no other documentation/forums you can search up, and even if you tried AI agents couldn’t do it.
Resource constrained environments are nice to develop for

2026-03-04 04:29:26

- reply

Years of old content already exists. At this point, it is set in stone.
Unlike centralized platforms there is no apply migration button

2026-03-04 04:27:34

- reply

On large closed models the FT possibilities are limited too.
These black boxes annoy me to no end, and there is little price competition.
They work good as teacher models for synthetic data, but even then I still use the -mini versions...

2026-03-04 04:22:17

- reply

attack surface *really* depends on the app, and usually people make it out to be way larger than it actually is normally

2026-03-03 23:44:18

- reply

I would only want either
- prompts / actual requests, or
- actual human made PRs that fit the style
no thanks I’d rather either write the code myself or run the agent myself

2026-03-03 23:27:21

- reply

for what benefit?

2026-03-03 23:25:04

- reply

Because no one cares about privacy, really.

2026-03-03 23:13:07

- reply

@npub14qt…9htj @npub1m3x…a5sf anything interesting on your side with these?

2026-03-03 23:11:43

oh well, a bunch of the top people at Qwen got replaced. ugh
we need a lot more innovation on open source *small* foundation models, as they will be what enable a lot of use cases, not your huge 1T parameter chatbot
#nevent1q…mjcu

2026-03-03 23:04:51

- reply

not planning to support in nfdb — it would
1. be useless for any relay with a client that has a wide-net filter
2. require a lot of global subscription state tracking, which I want to avoid
3. hold up the OK response until *all* other NFDB nodes ack the event

2026-03-03 23:02:43

- reply

last version my ass

2026-03-03 23:00:20

honestly the alby pivot doesn’t matter unless it actually affects the product’s quality.
don’t like shitcoins? disable it!
it’s very, very hard to make money in the bitcoin space, and the other option is just dying out instead of still being able to provide an extension

2026-03-03 22:23:39

- reply

fine tuning for actually useful tasks than a slop machine

2026-03-03 19:46:47

interesting, new small qwen 3.5 models with no instruction tuning

2026-03-02 23:54:38

- reply

also timestamp gate it probably

2026-03-02 23:52:05

- reply

Backwards compatibility

2026-03-02 20:56:44

- reply

in general do not quote unless:
1. you have something actually useful to add
2. are branching in another direction
3. want to make a more "fat" renote

2026-03-02 18:19:54

Does @npub1hyv…rtnu get any sleep? 😆

2026-03-02 18:17:55

- reply

Ah, the wonders of of early text-completion models and fine tuning BERT… 🥲

2026-03-02 16:41:25

- reply

it may be the specific type of codebase that makes it harder, but even with a ton of guidance, it takes much more effort to prompt than just doing it myself.
1-shot anything without guidance is stupid anyway
it is surprisingly good at writing tests + code reviews tho

2026-03-02 16:16:30

I do not get all of this AI coding hype still.
trying some of these tools and all they do is generate overly complex code that is unmaintainable, and they take longer to fix than to just do it right the 1st time around
https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/

2026-03-01 04:24:00

- reply

there are people trying to make businesses?
99.9% is incompetent people flinging sats at any random person on the street willing to pretend to work for <grant org>

2026-03-01 04:22:31

- reply

eh, it’s not nostr specific
next

2026-03-01 02:51:45

note: Nostr is not enshittification resistant. reject centralized “nostr” apps with dark patterns

2026-03-01 02:43:51

- reply

Would it not be possible to also do an app setting? I don’t want to see HDR in Damus but want to in other places.

2026-03-01 02:10:01

🤔
https://i.nostr.build/SsHJCtHYaYQSFxUk.png
https://i.nostr.build/CvwXL5LCpDjC1aGl.png

2026-03-01 00:34:31

- reply

@npub1xts…kk5s could you please fix hdr flashbangs in damus

2026-02-28 21:24:06

- reply

Fuck you, Bitcoin Shitty is better

2026-02-28 20:49:09

Both Knots and Core lost the plot

2026-02-28 20:48:06

- reply

I was going to post it an hour ago but I forgot 😅

2026-02-28 20:38:13

- reply

am I missing something? I want to see

2026-02-28 20:37:01

https://youtu.be/T4Upf_B9RLQ

2026-02-28 18:02:53

- reply

unrelated but I wouldn’t trust this site for too much technical details since it says his wallet was hacked *due to his pgp key*
it wasn’t, it was even worse: both on the same system for some reason??

2026-02-28 14:48:57

- reply

There is an adjacent initiative just about this that you might find useful and addresses many of the same things
https://www.sigstore.dev/

2026-02-28 14:43:00

- reply

Every package manager under the sun already does signature verification.

2026-02-27 22:46:59

- reply

My definition of bot spam is any bot that doesn’t follow these guidelines:
1. Don’t create more noise than necessary.
2. Do not interact unless explicitly authorized (such as by following WITH the user knowing, or explciitly tagged)
3. Minimize interaction scope if possible (untag irrelevant pubkeys)
4. Mark yourself as a bot and specify a human contact.

2026-02-26 15:55:03

- reply

Worst case make your own package repositories, not that hard. The primary thing that annoys me with package managers is no namespacing.
A package repository can override any package name. Which could lead to conflicts on independent sources

2026-02-26 15:41:07

I see all this vibe coding stuff going around on Nostr and yet there have been no new clients that weren’t slow asf or have horrible UX.

2026-02-26 10:12:23

- reply

None. LN is the common base layer and you’re just locking yourself into one badly-designed protocol instead of a standardlike NWC which works with eCash, custodial wallets, noncustodial wallets and anything LN based that doesn’t exist yet.

2026-02-25 23:35:33

- reply

That’s not for RAM but primarily batteries

2026-02-25 23:32:46

This RAM price crisis is insane.
It is close to the point where the RAM costs as much as the Linux eval board’s original price.
the bubble can’t pop sooner

2026-02-25 23:27:44

- reply

The only thing I slightly remember is something something HFT

2026-02-25 16:19:18

- reply

OSS only works if there is mutual benefit *and* collaboration
There is a reason NFDB is not open source

2026-02-24 21:00:42

- reply

oh, I am not the only one that thinks this?

2026-02-24 17:58:58

- reply

Clients can honor kind 5 deletions but they don’t.
It’s sloppiness from clients and an intentional disrespect of user privacy, not DRM.

2026-02-24 17:19:24

Why are devs so resistant to adding features to give users back control of their data?
This is one of the main things that annoys me, and that I am working on with the NFDB rewrite.
"You own your data" has become "Everyone owns your data, forever"
#nevent1q…rryr

2026-02-24 16:56:39

- reply

I will implement this when implementing new deletion features for new NFDB version. Thanks for the idea, unfortunate to see people do not want this.

2026-02-24 16:55:15

Wordle 1711 4/6
⬜⬜⬜🟩🟨
⬜🟨⬜⬜🟨
⬜⬜⬜🟨⬜
🟩🟩🟩🟩🟩

2026-02-24 16:28:04

- reply

or using a .png icon format, something standardized as a path maybe?

2026-02-24 16:26:54

- reply

🌍

2026-02-24 16:06:37

- reply

I should also say that writes to 1 relay can fail. So, outbox algorithm should read from multiple of users’ inboxes.

2026-02-24 15:55:03

- reply

👖