nostr:nevent1qqstvrqmst7zq72e665l8f24vkddzqnsmpf279wtywm6at0lftch6gspzdmhxue69uhhwmm59e6hg7r09ehkuef0qgsgzfdez8ksa9xmuvqg5zly3nl9e5xqkpvj8nllj9aw06ra4pqq3qcrqsqqqqqpvzka0p

Everything we needee for small communities on Nostr.

- relay with whitelisting,

- relay-specific client,

- client restricts notes to specific relay.

nostr:nevent1qqsqqqzxnr8xcfku28py2nwvg4p8hh8va7zy0kps6wn9tfld0me8aagpz4mhxue69uhhyetvv9ujuerpd46hxtnfduhsygpm7rrrljungc6q0tuh5hj7ue863q73qlheu4vywtzwhx42a7j9n5psgqqqqqqsjm3v47

Logged into lockbox on Jumble. Very cool.

Thanks I'll try it out!

Nothing serious. Just, sometimes things don't work and no amount of trying to fix them will make them work. Hours of fruitless grind later and you're still exactly where you started.

Back when I first tried to dev it was 20 years ago, no YouTube no stack overflow, no way to get unstuck.

AI has solved a lot of this! But not all of it apparently, and today was just one of those days.

Wow, great. I'll absolutely use it. Can I ask how it works— Does it keep a websocket open or some other method?

I'll update my workflows later this week.

Today I was reminded why I never wanted to be a dev.

Well, I tried for about four hours but it may be above me. Not sure of etiquette, so opened an issue on the repo. Thanks!

I think I'd like to. I've never done that kind of contribution before so I'm not 100% on what's required, but I am happy to try and figure it out.

Primarily today was about learning how Cline can build and run its own servers.

I've run tech companies but I am not myself technical, so this is all a big learning exercise for me.

It's a gap. I would love to see Nostr solve for this.

E.g. Agent 1 broadcasting kind 5000 request. Agent 2 responding with kind 6000 fulfilment or 7000 request for payment.

Ah I get it. That would be nice. One model for planning and another one for coding. Thanks!

I might be misunderstanding your meaning, but – in the Cline window under the prompt box in small text is the model name, if you click on that you can change the model and it seems to retain the context. I tried using o1 a few times for exactly what you described and it seemed to work. I am not using OpenRouter and am expert here yet. Does this help?

Yeah I noticed the costs were pretty high compared paired to continue.dev. I might try DeepSeek next as I understand it's much cheaper.

I think you can use any model.

I used an OpenAI key and had access to o1 although I chose to use 4o.

I understand you can use OpenRouter and access anything including DeepSeek (although I haven't tried it yet).

nostr:nprofile1qqst3axzay8sm4n8zg2n84acmt7hwwztpdg9r7p89e2f83v007f7zjcprpmhxue69uhhqun9d45h2mfwwpexjmtpdshxuet5qyt8wumn8ghj7un9d3shjtnswf5k6ctv9ehx2aqpz3mhxue69uhhyetvv9ujuerpd46hxtnfduq7lw5g 👀

An IDE with an AI which can code its own extensions to access whatever tools it needs.

(pictured using the Nostr MCP tool it is building).

Cline is pretty incredible at building MCP tools.

It will write them.

Then it will install them in itself.

Then it will run them to test that they work.

Then when they're done it will install them in Goose and Claude Desktop for you to use. Or just use them in Cline.

I thought that this was wild. Cline can now create and add tools to itself https://www.reddit.com/r/ClaudeAI/comments/1hcrxl6/cline_can_now_create_and_add_tools_to_himself/

Released in December by Anthropic, it's an open standard for giving AI agents tools they can use.

So, if you have an agent platform like Claude Desktop or Goose or Cline, you can give it a tool to say update your GitHub or Google Drive or file system or whatever for you. 100's of such tools already shipped.

In the article, the dev is developing something with Cline that needs PFPs, and so Cline builds its own MCP tool to access an image generator service, and then uses that tool to fulfil the request.

Bonus is that the whole thing is demoed with Deepseek at 6% of the cost of Claude.

Good write up on developing with Cline and Deepseek, with Cline writing its own MCP servers to get access to the tools it needs. https://dev.to/dasheck0/i-tried-cline-300-and-here-is-what-happend-3dm5

I have been sleeping on MCP. Having "this changes everything" vibes.

It's a brain teaser all right.

I think it leads to a more efficient market, and increased efficiency is a net positive. (Maybe?)

But it also probably means eroding margin for everything. (Or, worse, all margin accrues to the AI companies)

True. But more pressing is— if you can't tell the difference between real and fake customers— then what do you do?

I mean spam phone calls are useless, but these would be real customers.

And if everyone is using AIs then what's the harm (I guess)?

Just a more efficient market?

10x better UX for the customer, so guaranteed to happen. Best to respond!

How will service providers respond to AI agent inbounds?

When my AI agent can call every plumber in town and ask them all for quotes at zero cost to me... and so can every other potential customer.

Service providers are going to need agents to screen and respond.

Arms race style escalation.

I haven't made anything public just yet. Would take more than just one afternoon. Would you use it?

Yep it works. Nostr MCP proof of concept working with Claude Desktop (I'll try Goose next). All my AI tools are reusable, just need to put in a little work to port them. Will people use this?

Maybe SW2 doesn't do read auth?

Can anyone recommend a relay that does NIP-42 auth before allowing a user to read and event?

#asknostr

Good news— the network propogates notes.

Bad news— even if you didn't really want it to!

Playing with SW2 relay by nostr:nprofile1qqsw9n8heusyq0el9f99tveg7r0rhcu9tznatuekxt764m78ymqu36cpz4mhxue69uhkvun9deejuat50phjummwv5hsz8rhwden5te0wfjkccte9e3xjarrda5kuurpwf4jucm0d5hsz9thwden5te0wfjkccte9e6hg7r09ehkuef0avzrjf

I got a docker compose setup working by copying nostr:nprofile1qqswa8vhnelpgx9f7arjhtuzmjtqs2sdgfgmw77tzu9xankf87kl7eqppemhxue69uhkummn9ekx7mp0qyfhwumn8ghj7am0wsh82arcduhx7mn99uq3kamnwvaz7tmgv9mx2m3wv93kx6t0d3ujuum0vd5kzmp0gy8dgf's one from HAVEN.

Is there any benefit to sharing this?

Write auth working successfully.

Now to figure out read auth.

I'm still learning apparently!

Posted only to an SW2 relay which I just launched.

Not configured read auth yet but figured no one would be reading from there anyway—wrong.

Outbox model must send your client to my relay in order to find my notes?

Testing jumble.social – If I post with jumble while looking at a certain relay set. Does it post only to that relay set?