sandwich

MIkrotik is a decent option, RouterOS is pretty powerful, only gripe is that I've never had a smooth update. I have however only used it as a network combiner and as a failover, never as a full-fledged firewall.

2026-05-27 16:19:55 UTC

- reply

First, try opnsense instead.
Second, buy a dedicated box.
For the box, if you don't need a bunch of ethernet ports, Beelink is pretty good. otherwise if you have the budget check out Protectli and the official OPNSense product line; which has rack mounted options.

2026-05-27 11:49:20 UTC

- reply

you might need a whale to test that one

2026-05-27 09:04:24 UTC

- reply

lol 904kbps is absurd.

2026-05-26 19:29:37 UTC

- reply

Algos aren't inherently bad and never have been. The Reason why there is this perception is because the most influential social algos in recent history were the result of information asymmetry, jam-packed with hidden motives and for the benefit of unknown parties. When an algo is transparent and can be controlled by the same user it serves, an algorithm is greatly beneficial.

2026-05-26 19:00:53 UTC

- reply

But just in case he finds you, don't forget about Yhprum

2026-05-26 18:58:03 UTC

- reply

If that one gets you down, don't talk to Murphy.

2026-05-26 17:35:06 UTC

- reply

Snowcrash
The Unincorporated Man

2026-05-26 00:54:04 UTC

- reply

NGU is boring

2026-05-25 14:18:56 UTC

#asknostr
#nevent1q…37tx

2026-05-25 13:37:33 UTC

Is anyone here able to tell me the origin story of fractional reserve currencies, the reasons it was adopted, the problems it purported to solve and how the societal acceptance of its compromises led to central banks dominating international finance?

2026-05-25 10:56:37 UTC

- reply

Yes.

2026-05-25 00:09:32 UTC

- reply

NIP-05 is useful for validating in this situation

2026-05-24 23:28:52 UTC

Excellence survives adversity, it does not survive protection.
Self-sovereign individuals do whatever the fuck they want, and own all consequences.
Punk rock means refusing unearned authority while exercising independent judgment.
If an idea demands violence for its defense, it confesses its inability to endure scrutiny.
When character attack replaces engagement with the argument, no real refutation has occurred.
Chaos is the stable condition; indeterminism merely delays its resolution.

2026-05-24 18:03:52 UTC

- reply

if it's a static site with no SSR you can deploy as an nsite

2026-05-23 21:31:51 UTC

- reply

nah bro, nah.

2026-05-23 20:55:26 UTC

- reply

That's a bootstrapping problem, and DHT has that same problem.

2026-05-23 18:33:52 UTC

- reply

Tor is fine for nostr.

2026-05-23 17:40:13 UTC

- reply

It's already fairly resistent as it stands. If there were a DNS level attack, roughly 80% of relays would be affected, and almost all clients. Should be noted that there are around ~250 tor relays and i2p relays pop up every so often. You should also check out "nodns" and "epoxy" as well, they are different implementations that answer to this problem space.

2026-05-23 15:59:30 UTC

- reply

Small relays are already defacto supported by many outbox clients depending on their relay selection methodology, and local/cache relays are already supported by quite a few different applications.\

2026-05-23 15:38:44 UTC

- reply

DNS is currently nostr's lynchpin, by implementation rather than design. It's made worse by representing relays as URLls that are really just fuzzy strings that are practically impossible to validate or deduplicate.

2026-05-23 15:29:28 UTC

- reply

it's one of the survival mechanisms cognitive viruses use to spread themselves.

2026-05-23 15:24:48 UTC

- reply

Clarification because the message above is somewhat contradictory: I personally find no issue with a DHT when it exists above the protocol.

2026-05-23 15:21:48 UTC

- reply

It is a completely valid and technically sound solution. Should be noted that there are ways to circumvent DNS without a DHT; FIPS + pubkey relays effectively offloads the discovery mechanism without touching the protocol.

2026-05-23 15:15:03 UTC

- reply

I understand your position. Please consider the possibility that you may be underestimating the potential network effects that could result from implementing a DHT on nostr. Connecting to relays is the very first touch point of every application. It has the potential to short circuit and disincentivize the fundamental value proposition of the protocol.

2026-05-23 15:14:56 UTC

- reply

I believe you are mistaken. Most clients are outbox. The user is the network. A user opens a client, and the client finds pubkeys they follow, those pubkeys' relays, and starts pulling data cascading from the user outwards.

2026-05-23 14:46:41 UTC

- reply

DHT is an obvious solution for the perceived problem of data resolution, has been talked about since before I showed up in 2022 and is also why pubky exists to begin with; It's a better between technical accuracy and philosophical preference.
Something to consider you may not have already: what happens when clients start developing around the DHT as a first principle instead of NIP-01? There's a few implications of this at a high level:
1. Fewer and fewer clients will work in network segmented contexts because clients will assume all relays belong to a "global state," this limits the potential use cases of nostr. While there would be no technical limitation for clients to support non-DHT discovered relays, they would simply have no incentive to. Outbox would slowly die off, and the nostr ecosystem would be filled with applications that rely on global state for operation.
2. Social relationships are built outwards from the observer, nostr replicates this in its network topology. DHTs represent a global state and then a the observer works from the global state back to themselves. There could be many different DHTs, but then you have arbitrary boundaries that are difficult to cross. The former reflects of healthy human networks, while the latter more closely reflects the human networks we see in mainstream social media that is at the root of an uptick of mental illness. It's difficult to my discomfort with this one into words.
3. Potential slippery slope, once we introduce one global state, then why not put everything on a DHT? Pubkeys? Topics? Etc. It potentially erodes decentralization on a long enough time-scale, and at a certain point it hits a hard upper limit, challenging the "global" intent.
From a purely technical perspective, DHT makes sense and is the obvious solution to the perceived "problem." There are some scaling issues associated with it, but as long as the scope is tight enough the ceiling is pretty high, so on a technical level it passes the sniff test.
On a philosophical level it potentially signals a higher-level retraction the paradigm-shift that nostr has proposed. Nostr is not a payments network, it doesn't need a global state. Humans operate best and are most happy in a localized context, the global context is what is drives much of societal decay that we see today. Nostr works, it's hard to make it work, it has patterns that are intuitive and challenge conventional understanding of network design, but it works and in a way that is fundamentally different way than anything else.
For myself and many others, nostr's "no global" philosophy is a feature, not a bug.
Regardless of my critique, great work! Interesting exploration. It is inevitable someone would eventually dive into this. But not what I'm personally looking for in when establishing digital human connections. Unfortunately, I fear this is where we will end up because it's such an obvious optimization, and it's difficult to refute.

2026-05-23 11:38:13 UTC

- reply

@npub1der…xzpc

2026-05-23 10:08:22 UTC

- reply

And this is why popularity is not signal, but noise.

2026-05-22 22:53:08 UTC

- reply

Ahead of its time. Not given enough credit. Adversarial network design.

2026-05-22 22:51:54 UTC

- reply

Koo-bur-net-ease

2026-05-22 18:58:21 UTC

- reply

And not go completely off the rails immediately

2026-05-22 09:52:53 UTC

- reply

microkernels and fiat mining

2026-05-21 20:28:39 UTC

- reply

Yes.

2026-05-21 20:25:28 UTC

- reply

I unfortunately do not have a linux laptop right now, only run linux on my desktops.

2026-05-21 19:15:58 UTC

- reply

Is my favorite.

2026-05-21 14:19:37 UTC

- reply

Lol.

2026-05-21 01:00:07 UTC

- reply

This is THE question.

2026-05-21 00:59:26 UTC

- reply

Very cool. @nprofile…fsyl

2026-05-20 20:12:14 UTC

- reply

I agree, didn't say we didn't.

2026-05-20 15:56:38 UTC

- reply

It's not that people are dumb, it's that there is no emergent need, and social hasn't been a killer feature for longer than a decade, almost closer to 2 decades at this point.
nostr is inevitable, there will be a canon moment and it will become obvious to people. We are building a refuge for people who don't know they need it yet. Focus on the refuge, not NGU.

2026-05-20 10:40:06 UTC

- reply

sure, but not even claude can prevent it from misbehaving.

2026-05-20 09:56:21 UTC

- reply

throwing rocks at qemu

2026-05-20 08:52:28 UTC

- reply

I hope you're maintaining a dream diary through all this, I'm sure they're gold. Running away from unwanted UTXOs. Bourne Identity, Gigi edition.

2026-05-20 08:50:37 UTC

- reply

😳

2026-05-19 23:38:55 UTC

- reply

It's an extremely unpleasant experience that isn't necessarily aurral.

2026-05-19 14:27:23 UTC

- reply

2026-05-19 11:17:20 UTC

- reply

We've come full circle :D

2026-05-19 10:24:29 UTC

- reply

Gateways are a stepping stone to adoption, they don't require anyone to download anything, you can access them by clicking a link. Perfection is the enemy of good. End game was always a browser, OS or local resoluiton:
(Discussion from July 2025)
https://stacker.news/items/1016727

2026-05-19 10:13:12 UTC

- reply

That's generally what happens when an Administration leaves behind 20 years worth of military equipment to the Taliban, no?

2026-05-19 09:48:35 UTC

- reply

Lol, I shouldn't write notes when I'm tired. Re-read my root note this morning and laughed at myself, we totally misunderstood each other that entire thread and the fault is in my root note in the last sentence.
Target for the microkernal is physical RISC-V hardware. Think cyberdeck. The kernel itself is compiled RV64 machine code, not bytecode-interpreted. Heaviest ops are RustCrypto and it compiles down to rv64 instructions.
Bytecode only enters the picture if I get WASM napplets to a happy place that isn't retarded, and even then, all crypto operations are passed back to host imports.

2026-05-18 23:10:44 UTC

- reply

piss off everyone on both sides*

2026-05-18 23:10:07 UTC

- reply

I am not going to touch that topic right now, what I have to say would piss everyone on both sides off :D

2026-05-18 22:31:02 UTC

- reply

and then zap yourself

2026-05-18 22:26:14 UTC

- reply

haha, oh man, that's a battle for another day.

2026-05-18 19:41:30 UTC

- reply

lol, again not wrong. Live dangerously.
I do have a few VisionFive 2's but they are all busy ATM.

2026-05-18 19:39:00 UTC

- reply

Oh, sorry I didn't mean when compared to WASM, compared to other architectures.

2026-05-18 19:35:33 UTC

- reply

compute*

2026-05-18 19:34:44 UTC

- reply

You're not wrong, and I had similar thoughts when I started. Reason is I already had a good RISC-V boilerplate for for completely different project and so I took the plunge. RISC-V has clear disadvantages when it comes to performance, but also clear advantages when it comes to licensing and being a significantly more open platform. Crypto related operations are insanely expensive, but a computer module + daughter board with a hardware crypto solution could potentially resolve that.
Regardless, this is a proof of concept, conversation piece and exploration in general. I don't really expect this to become production. It has been very fun though :D

2026-05-18 19:06:03 UTC

- reply

NIP with updates: github.com/nostr-protocol/nips/pull/2303

2026-05-18 18:37:56 UTC

- reply

I see it.
#nevent1q…vcg2
https://github.com/nostr-protocol/nips/pull/2303

2026-05-18 18:35:04 UTC

While everyone was arguing about an experiment I built a nostr-native RISC-V microkernel that implements a variant of NIP-5D with a native transport. The applet pattern means that nostr applications can be written without in-depth knowledge of nostr protocol. It also means that the nostr connection, key management, crypto and optimization layers remain logically separated from the application layer. Unlike other similar solutions to date, few compromises are made: the application layer can be as creative as it wants and only has to focus on UX, while the shell handles the heavy lifting.
I now have two implementations of NIP-5D, one via a web-based shell that has a runtime and another via native shell on RISC-V embedded in a RUST microkernel; with API parity. Both will be released when they are released, I cannot afford that much time for nostr anymore, and I will only release them when they are ready. I will block out some time soon to refine a Nostr Applet possibility.

2026-05-18 13:47:49 UTC

- reply

a link to the tool would be nice :D

2026-05-17 15:04:23 UTC

build microkernels.

2026-05-16 09:30:40 UTC

- reply

Amaze. Amaze. Amaze.

2026-05-16 09:29:50 UTC

- reply

Not sure I agree. The internet had the same "dilemma," and what created the problems we are now trying to solve, was the fear of nobody using it and an unhealthy fixation on NGU.

2026-05-15 10:35:15 UTC

- reply

ugh, you're*

2026-05-15 10:35:06 UTC

- reply

"Your absolutely right, I should kill you. Target Locked"

2026-05-14 13:21:04 UTC

- reply

keep telling yourself that.

2026-05-11 18:00:22 UTC

- reply

This is one of the benefits of napplets btw: https://github.com/nostr-protocol/nips/pull/2303

2026-05-11 17:59:21 UTC

- reply

Someone woke up and chose violence :)

2026-05-10 16:01:45 UTC

- reply

Self-sovereign individuals put whatever tf they want on their pizza.

2026-05-10 12:29:09 UTC

- reply

Thank you for providing a logical argument in this thread.
I agree with everything except that it's an intentional attack on bitcoin, more than anything it seems misguided.

2026-05-09 17:49:15 UTC

- reply

Mine just started using "--" so it could get passed the linter, then a race started, I added "--" and then it started using "---"
we are presently at "-----".

2026-05-09 15:37:02 UTC

- reply

the way it was intended or the way people who pretend Gin isn't a word pronounce it?

2026-05-09 12:29:31 UTC

- reply

You are a fantastic son.

2026-05-09 12:06:08 UTC

- reply

Also, you of all people know that nsites could be resolved with nip05, or with named sites, or through a native interface locally without ever seeing a URL at all. Just needs someone to create it.
IMO When promoting outside of nostr, promoting 44billion and other implementations like it makes the most sense, when promoting inside of nostr to other devs promoting nsites makes the most sense because it's the most open ended.
We don't need app stores where we're going.

2026-05-09 12:00:37 UTC

- reply

(or tap)

2026-05-09 12:00:25 UTC

- reply

(or type)

2026-05-09 12:00:00 UTC

- reply

the majority of the population doesn't type urls, they click them.

2026-05-09 11:55:17 UTC

- reply

It's not just 44billion, it's every nsite gateway, of which there are many.

2026-05-08 13:25:08 UTC

- reply

depends on my mood
test bleeding edge like once a day:
pi: proposition is flexibility/control
dirac: proposition is token efficiency
for daily driving local models, and also as a backend for agentic applications via thin-wrapper:
OpenCode
For when I just need to burn through tokens for a long time and want to leverage subsidized providers:
Codex / Claude (in that order)

2026-05-08 09:41:26 UTC

- reply

For sure, but in a 12M context window you can fit an entire codebase + it's docs, so context management is no longer resolved by the model itself. The model weights could be tuned to the subject matter as opposed to the subject matter + the low-level contextual management of that subject matter; specialization vs generalization. Would this not lead to smaller models? I don't know about you, but a large unknown percentage my token usage revolves around context and the management of it for the aim of quality control, consistency and long-running autonomy.

2026-05-08 09:24:21 UTC

- reply

Well, there's solutions like subquadratic that's claiming 12M token context windows at 1/5th the (actual hard) cost of subsidized OpenAI/Anthropic "Max" plans. No telling if that's BS or not though until those gates are open.

2026-05-08 09:13:42 UTC

- reply

That is slowly starting to change fortunately. Smaller models are getting better and better, and researchers are getting more clever.

2026-05-08 09:12:33 UTC

- reply

price to performance ratio. GB10 is basically a developer platform, 20 TFLOPS, watts to TFLOP perf is on the lower end as well. RTX PRO 6000s are 125 TFLOPS, Max-Q variant is 120 TFLOPS but supports multi-gpu setups (so can scale). You ofc need to add some overhead for the rig: PSU, Mobo, Ram, CPU and need to plan correctly so you don't introduce any bottlenecks.

2026-05-08 08:40:51 UTC

- reply

https://nsite.run/#what

2026-05-08 08:40:09 UTC

- reply

GO the RTX PRO 6000 Workstation Max-Q route instead.

2026-05-07 17:51:48 UTC

- reply

Since:
#nevent1q…vm63

2026-05-07 14:22:30 UTC

- reply

have you considered deploying it as an nsite?

2026-05-07 14:20:23 UTC

- reply

LOVE IT! Really like that people don't need to actually be using the game to be included in it. Really cool. @npub18va…7nzf made a relay-centric game as well, but different. The one I am working on is not relay-centric; but only improving on it on occasion.

2026-05-07 14:00:57 UTC

- reply

The Federal Reserve

2026-05-07 13:59:56 UTC

- reply

indeed.

2026-05-07 13:49:25 UTC

- reply

working on something similar, have a link? would love to see it :D